Visualdsp++® 5.0 Linker and Utilities Manual

W 5.0 Linker and Utilities Manual (including the ADSP-BFxxx, ADSP-21xxx, ADSP-TSxxx)

Revision 3.5, January 2011 Part Number 82-000420-03

Analog Devices, Inc. One Technology Way Norwood, Mass. 02062-9106 a Copyright Information

Disclaimer

Analog Devices, Inc. reserves the right to change this product without prior notice. Information furnished by Analog Devices is believed to be accurate and reliable. However, no responsibility is assumed by Analog Devices for its use; nor for any infringement of patents or other rights of third parties which may result from its use. No license is granted by impli- cation or otherwise under the patent rights of Analog Devices, Inc.

Trademark and Service Mark Notice

The Analog Devices logo, Blackfin, EZ-KIT Lite, SHARC, TigerSHARC, and VisualDSP++ are registered trademarks of Analog Devices, Inc. All other brand and product names are trademarks or service marks of their respective owners. CONTENTS

PREFACE Purpose of This Manual ...... xix Intended Audience ...... xix Manual Contents ...... xx What’s New in This Manual ...... xxi Technical or Customer Support ...... xxii Supported Processors ...... xxiii Product Information ...... xxiii Analog Devices Web Site ...... xxiii VisualDSP++ Online Documentation ...... xxiv Technical Library CD ...... xxiv EngineerZone ...... xxv Social Networking Web Sites ...... xxv Notation Conventions ...... xxvi INTRODUCTION Software Development Flow ...... 1-2 Compiling and Assembling ...... 1-3 Inputs – C/C++ and Assembly Sources ...... 1-3

VisualDSP++ 5.0 Linker and Utilites Manual iii

Contents

Input Section Directives in Assembly Code ...... 1-4 Input Section Directives in C/C++ Source Files ...... 1-5 Linking ...... 1-7 Linker and Assembler Preprocessor ...... 1-8 Loading and Splitting ...... 1-10 LINKER Linker Operation ...... 2-3 Directing Linker Operation ...... 2-4 Linking Process Rules ...... 2-5 Linker Description File Overview ...... 2-6 Linking Environment for Windows ...... 2-7 Project Builds ...... 2-7 Expert Linker ...... 2-9 Linker Warning and Error Messages ...... 2-10 Link Target Description ...... 2-11 Representing Memory Architecture ...... 2-11 Specifying the Memory Map ...... 2-12 Memory Usage and Default Memory Segments ...... 2-12 Default Memory Segments for SHARC Processors ...... 2-14 Other Memory Segments ...... 2-18 Default Memory Segments for TigerSHARC Processors ..... 2-19 Other Memory Segments ...... 2-21 Default Memory Segments for Blackfin Processors ...... 2-21 Other Memory Segments ...... 2-24 iv VisualDSP++ 5.0 Linker and Utilites Manual

Contents

Blackfin Special “Table” Input Sections ...... 2-24 Input Sections in Blackfin Default LDFs for User Code/Data 2-26 Memory Characteristics Overview ...... 2-27 SHARC Memory Characteristics ...... 2-27 TigerSHARC Memory Characteristics ...... 2-30 Blackfin Memory Characteristics ...... 2-32 Linker MEMORY{} Command in an LDF ...... 2-32 Entry Address ...... 2-34 Wildcard Characters ...... 2-35 Placing Code on the Target ...... 2-36 Specifying Two Buffers in Different Memory Segments ...... 2-41 Linking with Attributes – Overview ...... 2-42 Profile-Guided Optimization Support ...... 2-43 Passing Arguments for Simulation or Emulation ...... 2-44 Linker Command-Line Reference ...... 2-44 Linker Command-Line Syntax ...... 2-45 Command-Line Object Files ...... 2-46 Command-Line File Names ...... 2-47 Object File Types ...... 2-49 Linker Command-Line Switches ...... 2-49 Linker Switch Summary and Descriptions ...... 2-51 @filename ...... 2-53 -Dprocessor ...... 2-53 -L path ...... 2-54

VisualDSP++ 5.0 Linker and Utilites Manual v

Contents

-M ...... 2-54 -MM ...... 2-54 -Map filename ...... 2-55 -MDmacro[=def] ...... 2-55 -MUDmacro ...... 2-56 -S ...... 2-56 -T filename ...... 2-56 -Werror [number] ...... 2-57 -Wwarn [number] ...... 2-57 -Wnumber[,number] ...... 2-57 -e ...... 2-57 -ek sectionName ...... 2-57 -es sectionName ...... 2-58 -entry ...... 2-58 -ev ...... 2-58 -flags-meminit -opt1[,-opt2...] ...... 2-58 -flags-pp-opt1[,-opt2...] ...... 2-58 -h[elp] ...... 2-59 -i|I directory ...... 2-59 -ip ...... 2-59 -jcs2l ...... 2-60 -jcs2l+ ...... 2-60 -keep symbolName ...... 2-60 -meminit ...... 2-61

vi VisualDSP++ 5.0 Linker and Utilites Manual

Contents

-nomemcheck ...... 2-61 -o filename ...... 2-61 -od directory ...... 2-61 -pp ...... 2-62 -proc processor ...... 2-62 -reserve-null ...... 2-62 -s ...... 2-62 -save-temps ...... 2-63 -si-revision version ...... 2-63 -sp ...... 2-64 -t ...... 2-64 -tx ...... 2-64 -v[erbose] ...... 2-64 -version ...... 2-64 -warnonce ...... 2-64 -xref ...... 2-65 LINKER DESCRIPTION FILE LDF File Overview ...... 3-3 Blackfin-Generated LDFs ...... 3-3 Default LDFs ...... 3-4 Example 1 – Basic LDF for Blackfin Processors ...... 3-7 Memory Usage in Blackfin Processors ...... 3-9 Example 2 – Basic LDF for TigerSHARC Processors ...... 3-10 Example 3 – Basic LDF for SHARC Processors ...... 3-11

VisualDSP++ 5.0 Linker and Utilites Manual vii

Contents

Common Notes on Basic LDF Examples ...... 3-13 LDF File Structure ...... 3-18 Command Scoping ...... 3-19 LDF Expressions ...... 3-20 LDF Keywords, Commands, and Operators ...... 3-21 LDF Keywords ...... 3-22 Miscellaneous LDF Keywords ...... 3-23 LDF Operators ...... 3-23 ABSOLUTE() Operator ...... 3-23 ADDR() Operator ...... 3-24 DEFINED() Operator ...... 3-26 MEMORY_END() Operator ...... 3-26 MEMORY_SIZEOF() Operator ...... 3-27 MEMORY_START() Operator ...... 3-27 SIZEOF() Operator ...... 3-28 Location Counter (.) ...... 3-29 LDF Macros ...... 3-29 Built-In LDF Macros ...... 3-30 User-Declared Macros ...... 3-32 LDF Macros and Command-Line Interaction ...... 3-32 Built-in Preprocessor Macros ...... 3-33 __VISUALDSPVERSION__ ...... 3-33 __VERSIONNUM__ ...... 3-35 __VERSION__ ...... 3-35

viii VisualDSP++ 5.0 Linker and Utilites Manual

Contents

__SILICON_REVISION__ ...... 3-36 __MEMINIT__ ...... 3-36 LDF Commands ...... 3-36 ALIGN() ...... 3-37 ARCHITECTURE() ...... 3-38 COMMON_MEMORY{} ...... 3-38 ELIMINATE() ...... 3-39 ELIMINATE_SECTIONS() ...... 3-40 ENTRY() ...... 3-40 INCLUDE() ...... 3-40 INPUT_SECTION_ALIGN() ...... 3-40 KEEP() ...... 3-42 KEEP_SECTIONS() ...... 3-42 LINK_AGAINST() ...... 3-42 MAP() ...... 3-43 MEMORY{} ...... 3-44 Segment Declarations ...... 3-45 segment_name ...... 3-45 START(address_number) ...... 3-45 TYPE() ...... 3-46 LENGTH(length_number)/END(address_number) ...... 3-47 WIDTH(width_number) ...... 3-47 MPMEMORY{} ...... 3-47 OVERLAY_GROUP{} ...... 3-48

VisualDSP++ 5.0 Linker and Utilites Manual ix

Contents

PACKING() ...... 3-48 Packing in SHARC Processors ...... 3-50 Overlay Packing Formats in SHARC Processors ...... 3-51 External Execution Packing in SHARC Processors ...... 3-52 PLIT{} ...... 3-53 PROCESSOR{} ...... 3-54 RESERVE() ...... 3-56 Linker Error Resolutions ...... 3-57 Example ...... 3-58 RESERVE_EXPAND() ...... 3-58 RESOLVE() ...... 3-59 Potential Problem with Symbol Definition ...... 3-59 SEARCH_DIR() ...... 3-60 SECTIONS{} ...... 3-61 INPUT_SECTIONS() ...... 3-64 Using an Optional Filter Expression ...... 3-65 INPUT_SECTIONS_PIN/_PIN_EXCLUSIVE Commands 3-67 expression ...... 3-69 FILL(hex number) ...... 3-69 PLIT{plit_commands} ...... 3-69 OVERLAY_INPUT{overlay_commands} ...... 3-70 FORCE_CONTIGUITY/NOFORCE_CONTIGUITY .... 3-72 SHARED_MEMORY{} ...... 3-72

x VisualDSP++ 5.0 Linker and Utilites Manual

Contents

EXPERT LINKER Expert Linker Overview ...... 4-2 Launching the Create LDF Wizard ...... 4-3 Step 1: Specifying Project Information ...... 4-5 Step 2: Specifying System Information ...... 4-6 Step 3: Completing the LDF Wizard ...... 4-8 Expert Linker Window Overview ...... 4-9 Input Sections Pane ...... 4-10 Input Sections Menu ...... 4-10 Mapping an Input Section to an Output Section ...... 4-12 Viewing Icons and Colors ...... 4-13 Sorting Objects ...... 4-15 Memory Map Pane ...... 4-16 Context Menu ...... 4-19 Tree View Memory Map Representation ...... 4-21 Graphical View Memory Map Representation ...... 4-22 Specifying Pre- and Post-Link Memory Map View ...... 4-26 Zooming In and Out on the Memory Map ...... 4-28 Adding a Memory Segment ...... 4-29 Inserting a Gap Into a Memory Segment ...... 4-31 Working With Overlays ...... 4-32 Viewing Section Contents ...... 4-33 Viewing Symbols ...... 4-36 Profiling Object Sections ...... 4-37

VisualDSP++ 5.0 Linker and Utilites Manual xi

Contents

Adding Shared Memory Segments and Linking Object Files ... 4-42 Managing Object Properties ...... 4-47 Managing General Global Properties ...... 4-48 Managing Processor Properties ...... 4-49 Managing PLIT Properties for Overlays ...... 4-50 Managing Elimination Properties ...... 4-51 Managing Symbols Properties ...... 4-53 Managing Memory Segment Properties ...... 4-57 Managing Output Section Properties ...... 4-58 Managing Packing Properties ...... 4-61 Managing Alignment and Fill Properties ...... 4-63 Managing Overlay Properties ...... 4-65 Managing Stack and Heap in Processor Memory ...... 4-67 Managing Shared Memory Properties ...... 4-70 MEMORY OVERLAYS AND ADVANCED LDF COMMANDS Overview ...... 5-2 Memory Management Using Overlays ...... 5-4 Introduction to Memory Overlays ...... 5-5 Overlay Managers ...... 5-7 Breakpoints on Overlays ...... 5-7 Memory Overlay Support ...... 5-8 Example – Managing Two Overlays ...... 5-13 Linker-Generated Constants ...... 5-15 xii VisualDSP++ 5.0 Linker and Utilites Manual

Contents

Overlay Word Sizes ...... 5-16 Storing Overlay ID ...... 5-20 Overlay Manager Function Summary ...... 5-20 Reducing Overlay Manager Overhead ...... 5-21 Using PLIT{} and Overlay Manager ...... 5-25 Inter-Overlay Calls ...... 5-27 Inter-Processor Calls ...... 5-28 Advanced LDF Commands ...... 5-29 OVERLAY_GROUP{} ...... 5-29 Ungrouped Overlay Execution ...... 5-31 Grouped Overlay Execution ...... 5-33 PLIT{} ...... 5-34 PLIT Syntax ...... 5-35 Command Evaluation and Setup ...... 5-36 Overlay PLIT Requirements and PLIT Examples ...... 5-36 PLIT – Summary ...... 5-38 Linking Multiprocessor Systems ...... 5-39 Selecting Code and Data for Placement ...... 5-40 Using LDF Macros for Placement ...... 5-40 Mapping by Section Name ...... 5-42 Mapping Using Attributes ...... 5-43 Mapping Using Archives ...... 5-44 MPMEMORY{} ...... 5-45 SHARED_MEMORY{} ...... 5-47

VisualDSP++ 5.0 Linker and Utilites Manual xiii

Contents

COMMON_MEMORY{} ...... 5-53 ARCHIVER Introduction ...... 6-2 Archiver Guide ...... 6-3 Creating a Library ...... 6-3 Making Archived Functions Usable ...... 6-4 Writing Archive Routines: Creating Entry Points ...... 6-4 Accessing Archived Functions From Your Code ...... 6-5 Specifying Object Files ...... 6-6 Tagging an Archive With Version Information ...... 6-7 Basic Version Information ...... 6-7 User-Defined Version Information ...... 6-8 Printing Version Information ...... 6-9 Removing Version Information From an Archive ...... 6-10 Checking Version Number ...... 6-10 Archiver Symbol Name Encryption ...... 6-10 Archiver Command-Line Reference ...... 6-14 elfar Command Syntax ...... 6-14 Archiver Parameters and Switches ...... 6-15 Command-Line Constraints ...... 6-17 MEMORY INITIALIZER Memory Initializer Overview ...... 7-2 Basic Operation of Memory Initializer ...... 7-3

xiv VisualDSP++ 5.0 Linker and Utilites Manual

Contents

Input and Output Files ...... 7-3 Initialization Stream Structure ...... 7-5 Run-Time Library Routine Basic Operation ...... 7-6 Using Memory Initializer ...... 7-7 Preparing the Linker Description File (.ldf) ...... 7-7 Preparing the Source Files ...... 7-9 Invoking Memory Initializer ...... 7-10 Invoking meminit from the VisualDSP++ IDDE ...... 7-10 Invoking meminit from the Command Line ...... 7-11 Invoking meminit from the Linker’s Command Line ...... 7-12 Invoking meminit from the Compiler’s Command Line ..... 7-12 Invoking meminit with Callback Executables ...... 7-12 Memory Initializer Command-Line Switches ...... 7-14 -BeginInit Initsymbol ...... 7-15 -h[elp] ...... 7-16 -IgnoreSection Sectionname ...... 7-16 -Init Initcode.dxe ...... 7-16 InputFile.dxe ...... 7-17 -NoAuto ...... 7-17 -NoErase ...... 7-17 -o Outputfile.dxe ...... 7-18 -Section Sectionname ...... 7-18 -v ...... 7-18

VisualDSP++ 5.0 Linker and Utilites Manual xv

Contents

FILE FORMATS Source Files ...... A-2 C/C++ Source Files ...... A-2 Assembly Source Files (.asm) ...... A-3 Assembly Initialization Data Files (.dat) ...... A-3 Header Files (.h) ...... A-4 Linker Description Files (.ldf) ...... A-4 Linker Command-Line Files (.txt) ...... A-5 Build Files ...... A-5 Assembler Object Files (.doj) ...... A-5 Library Files (.dlb) ...... A-6 Linker Output Files (.dxe, .sm, and .ovl) ...... A-6 Memory Map Files (.xml) ...... A-6 Loader Output Files in Intel Hex-32 Format (.ldr) ...... A-6 Splitter Output Files in ASCII Format (.ldr) ...... A-8 Debugger Files ...... A-9 Format References ...... A-10 UTILITIES elfdump – ELF File Dumper ...... B-1 Disassembling a Library Member ...... B-3 Dumping Overlay Library Files ...... B-4 elfpatch ...... B-5 Extracting a Section in an ELF File ...... B-5

xvi VisualDSP++ 5.0 Linker and Utilites Manual

Contents

Replacing Raw Contents of a Section in an ELF File ...... B-6 plinker ...... B-6 LDF PROGRAMMING EXAMPLES FOR BLACKFIN PROCESSORS Linking for a Single-Processor System ...... C-2 Linking Large Uninitialized or Zero-initialized Variables ...... C-4 LDF PROGRAMMING EXAMPLES FOR SHARC PROCESSORS Linking a Single-Processor SHARC System ...... D-2 Linking Large Uninitialized Variables ...... D-4 Linking for MP and Shared Memory ...... D-6 Reflective Semaphores ...... D-12 LDF PROGRAMMING EXAMPLES FOR TIGERSHARC PROCESSORS Linking a Single-Processor System ...... E-2 Linking Large Uninitialized or Zero-Initialized Variables ...... E-4 Linking an ADSP-TS101 MP Shared Memory System ...... E-6

VisualDSP++ 5.0 Linker and Utilites Manual xvii

Contents

xviii VisualDSP++ 5.0 Linker and Utilites Manual

PREFACE

Thank you for purchasing Analog Devices, Inc. development software for Analog Devices embedded processors.

Purpose of This Manual

The VisualDSP++ 5.0 Linker and Utilities Manual contains information about the linker and utility programs for Blackfin® (ADSP-BFxxx), TigerSHARC® (ADSP-TSxxx), and SHARC® (ADSP-21xxx) processors. These processors set a new standard of performance for digital signal processors, combining multiple computation units for floating-point and fixed-point processing as well as wide word width. The manual describes the linking process in the VisualDSP++ Windows application environment. This manual provides information on the linking process and describes the syntax for the linker’s command language—a scripting language that the linker reads from the linker description file (.ldf). The manual leads you through using the linker, archiver, and utilities to produce DSP programs and provides reference information on the file utility software.

Intended Audience

The primary audience for this manual is programmers familiar with Analog Devices processors. This manual assumes that the audience has a working knowledge of the appropriate processor architecture and instruction set.

VisualDSP++ 5.0 Linker and Utilities Manual xix Manual Contents

Programmers who are unfamiliar with Analog Devices processors can use this manual, but should supplement it with other texts (such as the appropriate Hardware Reference and Programming Reference manuals) that describe your target architecture.

Manual Contents

The manual contains: • Chapter 1, “Introduction”, provides an overview of the linker and utility programs. • Chapter 2, “Linker”, describes how to combine object files into reusable library files to link routines referenced by other object files. • Chapter 3, “Linker Description File”, describes how to write an .ldf file to define the target. • Chapter 4, “Expert Linker”, describes the Expert Linker, which is an interactive graphical tool for setting up and mapping processor memory. • Chapter 5, “Memory Overlays and Advanced LDF Commands”, describes how overlays and advanced LDF commands are used for memory management and complex linking.

• Chapter 6, “Archiver”, describes the elfar archiver utility used to combine object files into library files, which serve as reusable resources for code development. • Chapter 7, “Memory Initializer”, describes the Memory Initializer utility that is used to generate a single initialization stream and save it in a section in the output executable file.

xx VisualDSP++ 5.0 Linker and Utilities Manual Preface

• Appendix A, “File Formats”, lists and describes the file formats that the development tools use as inputs or produce as outputs. • Appendix B, “Utilities”, describes the utility programs that provide legacy and file conversion support. • Appendix C, “LDF Programming Examples for TigerSHARC Pro- cessors”, provides code examples of .ldf files for TigerSHARC processors • Appendix D, “LDF Programming Examples for SHARC Proces- sors”, provides code examples of .ldf files used with SHARC processors. • Appendix E, “LDF Programming Examples for Blackfin Proces- sors”, provides code examples of .ldf files used with Blackfin processors.

What’s New in This Manual

The VisualDSP++ 5.0 Linker and Utilities Manual documents linker support for all currently available Analog Devices’ SHARC, TigerSHARC and Blackfin processors. This edition includes modifications due to new processors and fixes to reported problems. Refer to VisualDSP++ 5.0 Product Release Bulletin for information on all new and updated VisualDSP++® 5.0 features and other release information.

VisualDSP++ 5.0 Linker and Utilities Manual xxi Technical or Customer Support

Technical or Customer Support

You can reach Analog Devices, Inc. Customer Support in the following ways: • Visit the Embedded Processing and DSP products Web site at http://www.analog.com/processors/technical_support

• E-mail tools questions to [email protected]

• E-mail processor questions to [email protected] (World wide support) [email protected] (Europe support) [email protected] (China support)

• Phone questions to 1-800-ANALOGD • Contact your Analog Devices, Inc. local sales office or authorized distributor • Send questions by mail to: Analog Devices, Inc. One Technology Way P.O. Box 9106 Norwood, MA 02062-9106 USA

xxii VisualDSP++ 5.0 Linker and Utilities Manual Preface

Supported Processors

This manual supports the following Analog Devices, Inc. processors. • Blackfin® (ADSP-BFxxx) •SHARC® (ADSP-21xxx) • TigerSHARC® (ADSP-TSxxx) The majority of the information in this manual applies to all processors. Information applicable to a particular target processor, or to a particular processor family, is provided in the appendices.

Product Information

Product information can be obtained from the Analog Devices Web site, VisualDSP++ online Help system, and a technical library CD.

Analog Devices Web Site

The Analog Devices Web site, www.analog.com, provides information about a broad range of products—analog integrated circuits, amplifiers, converters, and digital signal processors. To access a complete technical library for each processor family, go to http://www.analog.com/processors/technical_library. The manuals selection opens a list of current manuals related to the product as well as a link to the previous revisions of the manuals. When locating your manual title, note a possible errata check mark next to the title that leads to the current correction report against the manual.

Also note, MyAnalog.com is a free feature of the Analog Devices Web site that allows customization of a Web page to display only the latest infor-

VisualDSP++ 5.0 Linker and Utilities Manual xxiii Product Information

mation about products you are interested in. You can choose to receive weekly e-mail notifications containing updates to the Web pages that meet your interests, including documentation errata against all manuals. MyAna- log.com provides access to books, application notes, data sheets, code examples, and more.

Visit MyAnalog.com to sign up. If you are a registered user, just log on. Your user name is your e-mail address.

VisualDSP++ Online Documentation

Online documentation comprises the VisualDSP++ Help system, software tools manuals, hardware tools manuals, processor manuals, Dinkum Abridged C++ library, and FLEXnet License Tools documentation. You can search easily across the entire VisualDSP++ documentation set for any topic of interest.

For easy printing, supplementary Portable Documentation Format (.pdf) files for all manuals are provided on the VisualDSP++ installation CD. Each documentation file type is described as follows.

File Description

.chm Help system files and manuals in Microsoft help format

.htm or Dinkum Abridged C++ library and FLEXnet license tools software .html documentation. Viewing and printing the .html files requires a browser, such as Internet Explorer 6.0 (or higher).

.pdf VisualDSP++ and processor manuals in PDF format. Viewing and printing the .pdf files requires a PDF reader, such as Adobe Acrobat Reader (4.0 or higher).

Technical Library CD

The technical library CD contains seminar materials, product highlights, a selection guide, and documentation files of processor manuals, Visu- alDSP++ software manuals, and hardware tools manuals for the following xxiv VisualDSP++ 5.0 Linker and Utilities Manual Preface

processor families: Blackfin, SHARC, TigerSHARC, ADSP-218x, and ADSP-219x.

To order the technical library CD, go to http://www.analog.com/processors/technical_library, navigate to the manuals page for your processor, click the request CD check mark, and fill out the order form. Data sheets, which can be downloaded from the Analog Devices Web site, change rapidly, and therefore are not included on the technical library CD. Technical manuals change periodically. Check the Web site for the latest manual revisions and associated documentation errata.

EngineerZone

EngineerZone is a technical support forum from Analog Devices. It allows you direct access to ADI technical support engineers. You can search FAQs and technical information to get quick answers to your embedded processing and DSP design questions. Use EngineerZone to connect with other DSP developers who face similar design challenges. You can also use this open forum to share knowledge and collaborate with the ADI support team and your peers. Visit http://ez.analog.com to sign up.

Social Networking Web Sites

You can now follow Analog Devices processor development on Twitter and LinkedIn. To access:

• Twitter: http://twitter.com/ADIsharc and http://twitter.com/blackfin

• LinkedIn: Network with the LinkedIn group, Analog Devices SHARC or Analog Devices Blackfin: http://www.linkedin.com

VisualDSP++ 5.0 Linker and Utilities Manual xxv Notation Conventions

Notation Conventions

Text conventions used in this manual are identified and described as follows.

Example Description Close command Titles in bold style reference sections indicate the location of an item (File menu) within the VisualDSP++ environment’s menu system (for example, the Close command appears on the File menu).

{this | that} Alternative required items in syntax descriptions appear within curly brackets and separated by vertical bars; read the example as this or that. One or the other is required.

[this | that] Optional items in syntax descriptions appear within brackets and separated by vertical bars; read the example as an optional this or that.

[this,…] Optional item lists in syntax descriptions appear within brackets delimited by commas and terminated with an ellipse; read the example as an optional comma-separated list of this.

.SECTION Commands, directives, keywords, and feature names are in text with letter gothic font.

filename Non-keyword placeholders appear in text with italic style format.

Note: For correct operation, ... A Note provides supplementary information on a related topic. In the L online version of this book, the word Note appears instead of this symbol.

Caution: Incorrect device operation may result if ... Caution: Device damage may result if ... A Caution identifies conditions or inappropriate usage of the product a that could lead to undesirable results or product damage. In the online version of this book, the word Caution appears instead of this symbol. Warning: Injury to device users may result if ... A Warning identifies conditions or inappropriate usage of the product that could lead to conditions that are potentially hazardous for devices [ users. In the online version of this book, the word Warning appears instead of this symbol.

xxvi VisualDSP++ 5.0 Linker and Utilities Manual 1 INTRODUCTION

This chapter provides an overview of VisualDSP++ development tools and their use in the [DSP] project development process. The code examples in this manual have been compiled using L VisualDSP++ 5.0. The examples compiled with other versions of VisualDSP++ may result in build errors or different output although the highlighted algorithms stand and should continue to stand in future releases of VisualDSP++. This chapter includes: • “Software Development Flow” on page 1-2 • “Compiling and Assembling” on page 1-3 • “Linking” on page 1-7 • “Loading and Splitting” on page 1-10

VisualDSP++ 5.0 Linker and Utilities Manual 1-1 Software Development Flow

Software Development Flow

The majority of this manual describes linking, a critical stage in the program development process for embedded applications.

The linker tool (linker) consumes object and library files to produce executable files, which can be loaded onto a simulator or target processor. The linker also produces map files and other output that contain information used by the debugger. Debug information is embedded in the executable file. After running the linker, you test the output with a simulator or emulator. Refer to the VisualDSP++ User’s Guide and online Help for information about debugging. Finally, you process the debugged executable file(s) through the loader or splitter to create output for use on the actual processor. The output file may reside on another processor (host) or may be burned into a PROM. The VisualDSP++ 5.0 Loader and Utilities Manual describes loader/splitter functionality for the target processors. The processor software development flow can be split into three phases:

1. Compiling and assembling – Input source files C (.c), C++ (.cpp), and assembly (.asm) yield object files (.doj).

2. Linking – Under the direction of the linker description file (.ldf), a linker command line, and VisualDSP++ Project Options dialog box settings, the linker utility consumes object files (.doj) and library files (.dlb) to yield an executable (.dxe) file. If specified, shared memory (.sm) and overlay (.ovl) files are also produced.

3. Loading or splitting – The executable (.dxe) file, as well as shared memory (.sm) and overlay (.ovl) files, are processed to yield output file(s). For TigerSHARC and Blackfin processors, these are boot-loadable (.ldr) files or non-bootable PROM image files, which execute from the processor’s external memory.

1-2 VisualDSP++ 5.0 Linker and Utilities Manual Introduction

Compiling and Assembling

The process starts with source files written in C, C++, or assembly. The compiler (or a code developer who writes assembly code) organizes each distinct sequence of instructions or data into named sections, which become the main components acted upon by the linker.

Inputs – C/C++ and Assembly Sources

The first step toward producing an executable file is to compile or assemble C, C++, or assembly source files into object files. The VisualDSP++ development software assigns a .doj extension to object files (Figure 1-1).

Source Files (.c, .cpp, .asm) Object Files (.doj)

Compiler and Assembler

Figure 1-1. Compiling and Assembling

Object files produced by the compiler (via the assembler) and by the assembler itself consist of input sections. Each input section contains a particular type of compiled/assembled source code. For example, an input section may consist of program opcodes or data, such as variables of various widths. Some input sections may contain information to enable source-level debugging and other VisualDSP++ features. The linker maps each input section (via a corresponding output section in the executable) to a memory segment, a contiguous range of memory addresses on the target system.

Each input section in the .ldf file requires a unique name, as specified in the source code. Depending on whether the source is C, C++, or assembly, different conventions are used to name an input section (see “Linker Description File”).

VisualDSP++ 5.0 Linker and Utilities Manual 1-3 Compiling and Assembling

Input Section Directives in Assembly Code

A .SECTION directive defines a section in assembly source. This directive must precede its code or data. SHARC Code Example:

.SECTION/DM asmdata; // Declares section asmdata .VAR input[3]; // Declares data buffer in asmdata

.SECTION/PM asmcode; // Declares section asmcode R0 = 0x1234; // Three lines of code in asmcode R1 = 0x4567; R3 = R1 + R2;

In the above example, the /dm asmdata input section contains the array input, and the /pm asmcode input section contains the three line of code. Blackfin Code Example:

.SECTION Library_Code_Space; /* Section Directive */ .GLOBAL _abs; _abs: R0 = ABS R0; /* Take absolute value of input */ RTS; _abs.end;

In the above example, the assembler places the global symbol/label _abs and the code after the label into the input section Library_Code_Space, as it processes this file into object code. In the example, the linker knows what code is associated with the label _abs because it is delimited with the label _abs.end. For some linker features, especially unused section elimination (see “ELIMINATE_SECTIONS()” on page 3-40), the linker must be able to determine the end of code or data associated with a label. In assembly code, the end of a function data block can be marked with a label with the

1-4 VisualDSP++ 5.0 Linker and Utilities Manual Introduction

same name as the label at the start of the name with .end appended to it. It is also possible to prepend a “.” in which case the label will not appear in the symbol table which can make debugging easier.

Listing 1-1 shows uses of .end labels in assembly code.

Listing 1-1. Using Labels in Assembly Code

start_label: // code start_label.end // marks end of code section

new_label: // code new_label.END: // end label can be in upper case

one_entry: // function one_entry includes the code // in second_entry second_entry: // more code .one_entry.end: .second_entry.end: // prepended "." omits end label // from the symbol table

Input Section Directives in C/C++ Source Files

Typically, C/C++ code does not specify an input section name, so the compiler uses a default name. By default, the input section names are program (for code) and data1 (for data). Additional input section names are defined in .ldf files. (For more information on memory mapping, see “Specifying the Memory Map” on page 2-12.)

In C/C++ source files, you can use the optional section("name") C language extension to define sections. Example 1:

VisualDSP++ 5.0 Linker and Utilities Manual 1-5 Compiling and Assembling

While processing the following code, the compiler stores the temp variable in the ext_data input section of the .doj file and stores the code generated from func1 in an input section named extern.

... section ("ext_data") int temp; /* Section directive */ section ("extern") void func1(void) { int x = 1; } ...

Example 2:

The section ("name") extension is optional and applies only to the declaration to which it is applied. Note that the new function (func2) does not have section ("extern") and will be placed in the default input section program. For more information on LDF sections, refer to “Specifying the Memory Map” on page 2-12.

section ("ext_data") int temp; section ("extern") void func1(void) { int x = 1; } int func2(void) { return 13; } /* New */

For information on compiler default section names, refer to the VisualDSP++ 5.0 C/C++ Compiler and Library Manual for the appropriate target processor and “Placing Code on the Target” on page 2-36. Identify the difference between input section names, output sec- L tion names, and memory segment names because these types of names appear in the .ldf file. Usually, default names are used. However, in some situations you may want to use non-default names. One such situation is when various functions or variables (in the same source file) are to be placed into different memory segments.

1-6 VisualDSP++ 5.0 Linker and Utilities Manual Introduction

Linking

After you have (compiled and) assembled source files into object files, use the linker to combine the object files into an executable file. By default, the development software gives executable files a .dxe extension (Figure 1-2).

Library Files (.dlb)

Object Files (.doj) Executables (.dxe, .sm, .ovl)

Linker

Linker Description Project Options File (LDF) Dialog Box Settings

Figure 1-2. Linking Diagram

Linking enables your code to run efficiently in the target environment. Linking is described in detail in Chapter 3, “Linker”. When developing a new project, use the Project Wizard (Blackfin) L or Expert Linker (SHARC and TigerSHARC) to generate the project’s .ldf file. For more information, see Chapter 4, “Expert Linker” or search online help for “Project Wizard”.

VisualDSP++ 5.0 Linker and Utilities Manual 1-7 Linking

Linker and Assembler Preprocessor

The linker and assembler preprocessor program (pp.exe) evaluates and processes preprocessor commands in source files. With these commands, you direct the preprocessor to define macros and symbolic constants, include header files, test for errors, and control conditional assembly and compilation.

The pp preprocessor is run by the assembler or linker from the operating system’s command line or from within the VisualDSP++ environment. These tools accept and pass this command information to the preprocessor. The preprocessor can also operate from the command line using its own command-line switches. “.” Character Identifier

The assembler/linker preprocessor treats the “.” character as part of an identifier.

The preprocessor matches the assembler which uses “.” as part of assembler directives and as a valid character in labels. This behavior creates a possible problem for users that have written preprocessor macros that rely on identifiers to break when encountering the “.” character, usually seen when processing register names. For example,

#define Loadd(reg, val) \

reg.l = val; \

reg.h = val;

The above example would not work in VisualDSP++ 5.0 because Visu- alDSP++ 5.0 does not provide any replacement since reg is not parsed as a separate identifier. The macro must be rewritten using the ## operator, such as:

#define Loadd(reg, val) \

1-8 VisualDSP++ 5.0 Linker and Utilities Manual Introduction

reg ## .l = val; \

reg ## .h = val;

The preprocessor supports ANSI C standard preprocessing with L extensions but differs from the ANSI C standard preprocessor in several ways. For information on the pp preprocessor, see the VisualDSP++ 5.0 Assembler and Preprocessor Manual. The compiler has it own preprocessor that permits the use of L preprocessor commands within C/C++ source. The compiler preprocessor automatically runs before the compiler. For more information, see the VisualDSP++ 5.0 C/C++ Compiler and Library Manual for the appropriate target architecture.

VisualDSP++ 5.0 Linker and Utilities Manual 1-9 Loading and Splitting

Loading and Splitting

After debugging the .dxe file, you process it through a loader or splitter to create output files used by the actual processor. The file(s) may reside on another processor (host) or may be burned into a PROM. For more information, refer to the VisualDSP++ 5.0 Loader and Utilities Manual which provides detailed descriptions of the processes and options used to generate boot-loadable loader (.ldr) files for the appropriate target processor. This manual also describes the splitting utility, which creates the non-boot loadable files that execute from the processor’s external memory. In general: • SHARC ADSP-2106x/ADSP-21160 processors use the loader (elfloader.exe) to yield a boot-loadable image (.ldr file), which resides in memory external to the processor (PROM or host processor). Use the splitter utility (elfspl21k) to generate non-bootable PROM image files, which execute from the processor’s external memory (often used with the ADSP-21065L processors). • SHARC ADSP-2116x/2126x/2136x/2137x/2147x/2148x processors use the loader (elfloader.exe) to yield a boot-loadable image (.ldr file), which transported to (and run from) processor memory. To make a loadable file, the loader processes data from a boot-kernel file (.dxe) and one or more other executable files (.dxe).

• TigerSHARC processors use the loader (elfloader.exe) to yield a boot-loadable image (.ldr file), which is transported to (and run from) processor memory. To make a loadable file, the loader processes data from a boot-kernel file (.dxe) and one or more other executable files (.dxe).

1-10 VisualDSP++ 5.0 Linker and Utilities Manual Introduction

• TigerSHARC and SHARC processors use the splitter utility (elfspl21k.exe) to generate non-bootable PROM image files, which execute from the processor’s external memory.

• Blackfin processors use the loader (elfloader.exe) to yield a boot-loadable image (.ldr file), which resides in memory external to the processor (PROM or host processor. To make a loadable file, the loader processes data from a boot-kernel file (.dxe) and one or more other executable files (.dxe). Figure 1-3 shows a simple application of the loader. In this example, the loader’s input is a single executable (.dxe) file. The loader can accommodate up to two .dxe files as input plus one boot kernel file (.dxe).

Executables (.dxe, .sm, .ovl) Debugger (Simulator, ICE, or EZ-KIT Lite)

Loader

Boot Image (.ldr)

Boot Kernel (.dxe)

Figure 1-3. Using the Loader to Create an Output File

For example, when a TigerSHARC processor is reset, the boot kernel portion of the image is transferred to the processor’s core. Then, the

VisualDSP++ 5.0 Linker and Utilities Manual 1-11 Loading and Splitting

instruction and data portion of the image are loaded into the processor’s internal RAM (as shown in Figure 1-4) by the boot kernel.

EPROMEPROM ADSP-21 ProcessorDSP

11 BootBoot Kernel InternalInternal MemoryMemory 22

InstructionsInstructions andand DataData

Figure 1-4. Booting from a Bootloadable (.LDR) File

VisualDSP++ includes boot kernel files (.dxe), which are used automatically when you run the loader. You can also customize boot kernel source files (included with VisualDSP++) by modifying and rebuilding them. Figure 1-5 shows how multiple input files—in this case, two executable (.dxe) files, a shared memory (.sm) file, and overlay (.ovl) files—are consumed by the loader to create a single image file (.ldr). This example illustrates the generation of a loader file for a multiprocessor architecture.

1-12 VisualDSP++ 5.0 Linker and Utilities Manual Introduction

The .sm and .ovl files should reside in the same directory that L contains the input .dxe file(s) or in the current working directory. If your system does not use shared memory or overlays, .sm and .ovl files are not required.

1.ovl

2.ovl

1.dxe 2.dxe

1.sm 2.sm N.ovl

Loader

.ldr

Figure 1-5. Input Files for a Multiprocessor System

This example has two executable files that share memory. Overlays are also included. The resulting output is a compilation of all the inputs.

VisualDSP++ 5.0 Linker and Utilities Manual 1-13 Loading and Splitting

1-14 VisualDSP++ 5.0 Linker and Utilities Manual 2LINKER

Linking assigns code and data to processor memory. For a simple single processor architecture, a single .dxe file is generated. A single invocation of the linker may create multiple executable (.dxe) files for multiprocessor (MP) or multi-core (MC) architectures. Linking can also produce a shared memory (.sm) file for an MP or MC system. A large executable file can be split into a smaller executable file and overlay (.ovl) files, which contain code that is called in (swapped into internal processor memory) as needed. The linker performs this task. You can run the linker from a command line or from the VisualDSP++ Integrated Development and Debugging Environment (IDDE). You can load linker output into the VisualDSP++ debugger for simulation, testing, and profiling. This chapter includes: • “Linker Operation” on page 2-3 • “Linking Environment for Windows” on page 2-7 • “Linker Warning and Error Messages” on page 2-10 • “Link Target Description” on page 2-11 • “Linker Command-Line Reference” on page 2-44

VisualDSP++ 5.0 Linker and Utilities Manual 2-2 Linker Operation

Linker Operation

Figure 2-1 illustrates a basic linking operation. The figure shows several object (.doj) files being linked into a single executable (.dxe) file. The linker description file (.ldf) directs the linking process.

1.doj 2.doj n.doj .ldf

Linker

.dxe

Figure 2-1. Linking Object Files to Produce an Executable File

When developing a new project, use the Project Wizard (Blackfin) L or Expert Linker (SHARC and TigerSHARC) to generate the project’s LDF. For more information, see Chapter 4, “Expert Linker” or search online help for “Project Wizard”.

In a multiprocessor system, a .dxe file for each processor is generated. For example, for a dual-processor system, you must generate two .dxe files. The processors in a multiprocessor architecture may share memory. When directed by statements in the .ldf file, the linker produce a shared memory (.sm) executable file whose code is used by multiple processors. Overlay files, another linker output, support applications that require more program instructions and data than the processor’s internal memory can accommodate. Refer to “Memory Management Using Overlays” on page 5-4 for more information.

2-3 VisualDSP++ 5.0 Linker and Utilities Manual Linker

Similar to object files, executable files are partitioned into output sections with unique names. Output sections are defined by the Executable and Linking Format (ELF) file standard to which VisualDSP++ conforms. The executable’s input section names and output section names L occupy different namespaces. Because the namespaces are independent, the same section names may be used. The linker uses input section names as labels to locate corresponding input sections within object files.

The executable file(s) (.dxe) and auxiliary files (.sm and .ovl) are not loaded into the processor or burned onto an EPROM. These files are used to debug the application.

Directing Linker Operation

Linker operations are directed by these options and commands: • Linker command-line switches (options). Refer to “Linker Com- mand-Line Reference” on page 2-44. • In an IDDE environment: Options on the Link page of the Project Options dialog box. See “Project Builds” on page 2-7. • LDF commands. Refer to “LDF Commands” on page 3-36 for a detailed description. Linker options control how the linker processes object files and library files. These options specify various criteria such as search directories, map file output, and dead code elimination.

LDF commands in a linker description file (.ldf) define the target memory map and the placement of program sections within processor memory. The text of these commands provides the information needed to link your code.

VisualDSP++ 5.0 Linker and Utilities Manual 2-4 Linker Operation

The VisualDSP++ Project window displays the .ldf file as a source L file, though the file provides linker command input.

Using directives in the .ldf file, the linker: • Reads input sections in the object files and maps them to output sections in the executable file. More than one input section may be placed in an output section. • Maps each output section in the executable to a memory segment, a contiguous range of memory addresses on the target processor. More than one output section may be placed in a single memory segment.

Linking Process Rules

The linking process observes these rules: • Each source file produces one object file. • Source files may specify one or more input sections as destinations for compiled/assembled object(s). • The compiler and assembler produce object code with labels (input section names) that can be used to direct one or more portions of object code to particular input sections.

• As directed by the .ldf file, the linker maps each input section in the object code to an output section.

• As directed by the .ldf file, the linker maps each output section to a memory segment. • Each input section may contain multiple code items, but a code item may appear in one input section only. • More than one input section may be placed in an output section.

2-5 VisualDSP++ 5.0 Linker and Utilities Manual Linker

• Each memory segment must have a specified width. • Contiguous addresses on different-width hardware must reside in different memory segments. • More than one output section may map to a memory segment if the output sections fit completely within the memory segment.

Linker Description File Overview

Whether you are linking C/C++ functions or assembly routines, the mech- anism is the same. After converting the source files into object files, the linker uses directives in an .ldf file to combine the objects into an executable (.dxe) file, which may be loaded into a simulator for testing. Executable file structure conforms to the Executable and Linkable L Format (ELF) standard.

Each project must include one .ldf file that specifies the linking process by defining the target memory and mapping the code and data into that memory. You can write your own .ldf file, or you can modify an existing file; modification is often the easier alternative when there are few changes in your system’s hardware or software. VisualDSP++ provides an .ldf file that supports the default mapping of each processor type. When developing a new project, use the Project Wizard (Blackfin) L or Expert Linker (SHARC and TigerSHARC) to generate the project’s LDF. For more information, see Chapter 4, “Expert Linker” or search online help for “Project Wizard”.

Similar to an object (.doj) file, an executable (.dxe) file consists of different segments, called output sections. Input section names are independent of output section names. Because they exist in different namespaces, input section names can be the same as output section names. Refer to Chapter 3, “Linker Description File” for further information.

VisualDSP++ 5.0 Linker and Utilities Manual 2-6 Linking Environment for Windows

Linking Environment for Windows

The linking environment refers to Windows command-prompt windows and the VisualDSP++ IDDE. At a minimum, run development tools (such as the linker) via a command line and view output in standard output. VisualDSP++ provides an environment that simplifies the processor program build process. From VisualDSP++, you specify build options from the Project Options dialog box and modify files, including the linker description file (.ldf). The Project Options dialog box’s Type option allows you to choose whether to build a library (.dlb) file, an executable (.dxe) file, or an image file (.ldr or others). Error and warning messages appear in the Output window.

Project Builds

The linker runs from an operating system command line, issued from the VisualDSP++ IDDE or a command prompt window. The VisualDSP++ IDDE provides an intuitive interface for processor programming. When you open VisualDSP++, a work area contains everything needed to build, manage, and debug a DSP project. You can easily create or edit an .ldf file, which maps code or data to specific memory segments on the target. For information about the VisualDSP++ environment, refer to the L VisualDSP++ User’s Guide or online Help. Online Help provides powerful search capabilities. To obtain information on a code item, parameter, or error, select text in an VisualDSP++ IDDE editor window or Output window and press the keyboard’s F1 key. Within VisualDSP++, specify tool settings for project builds. Use the Project menu to open the Project Options dialog box. The dialog box pages allow you to select the target processor, type, and name of the executable file, as well as VisualDSP++ tools available for use with the selected processor.

2-7 VisualDSP++ 5.0 Linker and Utilities Manual Linker

When using the VisualDSP++ IDDE, use the Link page from the Project Options dialog box to select and/or set linker functional options.

Figure 2-2. Project Options – Link: General Page

There are four sub-pages you can access—General, LDF Preprocessing, Elimination, and Processor. Figure 2-2 shows a sample Project:Link:Gen- eral sub-page. Most dialog box options have a corresponding compiler command-line switch as described in “Linker Command-Line Switches” on page 2-49. Use the Additional options field on each sub-page to enter appropriate file names, switches, and parameters that do not have corresponding controls on the dialog box but are available as compiler switches. Due to different processor architectures, different Link page options are available. Use context-sensitive online Help in VisualDSP++ to obtain information on dialog box controls (linker options). To do so, click on the

VisualDSP++ 5.0 Linker and Utilities Manual 2-8 Linking Environment for Windows

“?” button and then click on the field, box, or button for which you need information.

Expert Linker

The VisualDSP++ IDDE provides an interactive tool, Expert Linker, to map code or data to specific memory segments. When developing a new project, use the Expert Linker to generate the LDF.

Windows-hosted Expert Linker graphically displays the .ldf information (object files, LDF macros, libraries, and a target memory description). With Expert Linker, use drag-and-drop operations to arrange the object files in a graphical memory mapping representation. When you are satisfied with the memory layout, generate the executable (.dxe) file. Figure 2-3 shows the Expert Linker window, which comprises two panes: Input Sections and Memory Map (output sections). Refer to Chapter 4, “Expert Linker” for detailed information.

Figure 2-3. Expert Linker Window

2-9 VisualDSP++ 5.0 Linker and Utilities Manual Linker

Linker Warning and Error Messages

Linker messages are written to the VisualDSP++ Output window or to standard output (when the linker is run from a command line). Messages describe problems the linker encountered while processing the .ldf file. Warnings indicate processing errors that do not prevent the linker from producing a valid output file, such as unused symbols in your code. Errors are issued when the linker encounters situations that prevent the production of a valid output file.

Typically, these messages include the name of the .ldf file, the line number containing the message, a six-character code, and a brief description of the condition. For example,

linker -proc ADSP-unknown a.doj [Error li1010] The processor ‘ADSP-unknown’ is unknown or unsupported.

Interpreting Linker Messages You can access descriptions of linker messages by selecting the six-character code (for example, li1010) and pressing the F1 key. Within VisualDSP++, the Output window’s Build page displays project build status and error messages. In most cases, double-clicking a message displays the line in the source file causing the problem. Some build errors, such as a reference to an undefined symbol, do not correlate directly to source files. These errors often stem from omissions in the .ldf file. For example, if an input section from the object file is not placed by the .ldf file, a cross-reference error occurs at every object that refers to labels in the missing section. Fix this problem by reviewing the .ldf file and specifying all sections that need placement. For more information, refer to online Help.

VisualDSP++ 5.0 Linker and Utilities Manual 2-10 Link Target Description

Link Target Description

Before defining the system’s memory and program placement with linker commands, analyze the target system to ensure you can describe the target in terms the linker can process. Then, produce an .ldf file for your project to specify these system attributes: • Physical memory map • Program placement within the system’s memory map

If the project does not include an .ldf file, the linker uses a default L .ldf file for the processor that matches the -proc switch on the linker’s command line (or the Processor selection specified on the Project page of the Project Options dialog box in the VisualDSP++ IDDE). Be sure to understand the processor’s memory architecture, which is described in the appropriate processor’s Hardware Reference and in its data sheet. This section contains: • “Representing Memory Architecture” on page 2-11 • “Specifying the Memory Map” on page 2-12 • “Placing Code on the Target” on page 2-36 • “Profile-Guided Optimization Support” on page 2-43 • “Passing Arguments for Simulation or Emulation” on page 2-44

Representing Memory Architecture

The .ldf file’s MEMORY{} command is used to represent the memory architecture of your processor system. The linker uses this information to place the executable file into the system’s memory.

2-11 VisualDSP++ 5.0 Linker and Utilities Manual Linker

Perform the following tasks to write a MEMORY{} command: • Memory Usage – List the ways your program uses memory in your system. Typical uses for memory segments include interrupt tables, initialization data, program code, data, heap space, and stack space. Refer to “Specifying the Memory Map” on page 2-12. • Memory Characteristics – List the types of memory in your processor system and the address ranges and word width associated with each memory type. Memory type is defined as RAM or ROM.

• MEMORY{} Command – Construct a MEMORY{} command to combine the information from the previous two lists and to declare your system’s memory segments. For complete information, refer to “MEMORY{}” on page 3-44.

Specifying the Memory Map

An embedded program must conform to the constraints imposed by the processor’s data path (bus) widths and addressing capabilities. The following information describes an .ldf file for a hypothetical project. This file specifies several memory segments that support the SECTIONS{} command, as shown in “SECTIONS{}” on page 3-61. The following topics are important when allocating memory: • “Memory Usage and Default Memory Segments” on page 2-12 • “Memory Characteristics Overview” on page 2-27 • “Linker MEMORY{} Command in an LDF” on page 2-32

Memory Usage and Default Memory Segments Input section names are generated automatically by the compiler or are specified in the assembly source code. The .ldf file defines memory seg-

VisualDSP++ 5.0 Linker and Utilities Manual 2-12 Link Target Description

ment names and output section names. The default .ldf file handles all compiler-generated input sections (refer to the “Input Section” column in Table 2-1, Table 2-2, and Table 2-3). The produced .dxe file has a corresponding output section for each input section. Although programmers typically do not use output section labels, the labels are used by down- stream tools.

Use the ELF file dumper utility (elfdump) to dump contents of an output section (for example, data1) of an executable file. See “elfdump – ELF File Dumper” on page B-1 for information about this utility. The following sections show how input sections, output sections, and memory segments correspond in the default .ldf files for the appropriate target processor.

Refer to your processor’s default .ldf file and to the processor’s L Hardware Reference for details. Also see “Wildcard Characters” on page 2-35. Typical uses for memory segments include interrupt tables, initialization data, program code, data, heap space, and stack space. For detailed processor-specific information, refer to: • “Default Memory Segments for SHARC Processors” • “Default Memory Segments for TigerSHARC Processors” • “Default Memory Segments for Blackfin Processors” • “Blackfin Special “Table” Input Sections”

2-13 VisualDSP++ 5.0 Linker and Utilities Manual Linker

Default Memory Segments for SHARC Processors

Table 2-1 shows section mapping in the default .ldf file for an ADSP-21161 processor (as a simplified example for SHARC processors)

Table 2-1. Section Mapping in the Default SHARC LDF File

Input Section Output Section Memory Segment

seg_pmco seg_pmco seg_pmco

seg_dmda seg_dmda seg_dmda

seg_pmda seg_pmda seg_pmda

seg_rth seg_rth seg_rth

seg_init seg_init seg_init

seg_init_code seg_init_code seg_init_code

seg_argv seg_argv seg_argv

seg_ctdm dxe_ctdm mem_ctdm seg_ctdml

seg_vtbl seg_vtbl seg_dmda

seg_sram seg_sram seg_sram

.bss .bss seg_dmda

.gdt seg_dmda seg_dmda .gdtl

.frt seg_dmda seg_dmda

.cht seg_dmda seg_dmda

.edt seg_dmda seg_dmda

.rtti seg_dmda seg_dmda VDK Only:

seg_stack stackseg seg_stack For ADSP-213xx/ADSP-214xx Processors Only:

seg_stak stackseg seg_stak

seg_ext_code seg_ext_code seg_ext_code

VisualDSP++ 5.0 Linker and Utilities Manual 2-14 Link Target Description

Table 2-1. Section Mapping in the Default SHARC LDF File (Cont’d)

Input Section Output Section Memory Segment

seg_heap heap seg_heap

seg_ext_data seg_ext_data seg_ext_dmda

seg_sdram seg_sdram_data seg_ext_dmda

seg_flash seg_flash seg_flash For ADSP-214xx Processors Only:

seg_ext_code seg_ext_code seg_ext_swco

seg_swco seg_swco seg_int_code

For more information on stack and heap allocation, see “Memory Usage” in the VisualDSP++ C/C++ Compiler Manual for SHARC Processors. Several input sections and memory segments are used in the default .ldf files for ADSP-210xx/211xx/212xx/213xx/214xx processors, which must be present in the user’s .ldf file. These sections are described in detail below. .bss This section contains global zero-initialized data. The linker places the contents of this data section in seg_dmda. .rtti This section is used by the C++ run-time type identification support, when enabled. seg_rth This section contains the interrupt vector table. By default, this is located in the start-up file (for example, 060_hdr.doj). seg_init This section contains location and size information about the stack and heap; also contains compressed data created by the memory initialization tool. (See “-meminit” on page 2-61 for more information.)

2-15 VisualDSP++ 5.0 Linker and Utilities Manual Linker

seg_int_code Code that modifies interrupt latch registers must not be executed from external memory. To minimize the impact of this restriction, the library functions that modify the latch registers are located in the seg_init_code section, which should be located in internal memory. seg_pmco This section is the default location for program code. seg_pmda This section is the default location for global program data that is qualified with the “pm” keyword. For example,

int pm xyz[100]; // Located in seg_pmda

seg_argv This section contains the command-line arguments that are used as part of profile-guided optimization (PGO). seg_ctdm This section contains the addresses of constructors called before the start of a C++ program (such as constructors for global and static objects). This section must be terminated with the symbol “___ctor_NULL_marker” (the default .ldf files ensure this). It is required if compiling with C++ code. seg_dmda This section is the default location for global data and for data that is qualified with the “dm” keyword. For example,

int abc[100]; // Located in seg_dmda

int dm def[100]; // Located in seg_dmda

In the default (non-VDK) LDFs for the ADSP-21020/2106x/2116x and ADSP-2126x processors, the run-time stack and heap are also allocated from this section.

VisualDSP++ 5.0 Linker and Utilities Manual 2-16 Link Target Description

seg_stack (VDK only) The run-time stack is located in this section. Local variables, function parameters, and so on are stored here. seg_stak (not VDK) In the LDFs for ADSP-2136x/2137x and ADSP-2147x/2148x processors, this section is the area where the run-time stack is located. Local variables, function parameters, and so on are stored here. In the LDFs for ADSP-21020, ADSP-2106x, ADSP-2116x, and ADSP-2126x processors, the run-time stack is located in seg_dmda. seg_vtbl This section contains C++ virtual function tables. The default .ldf files place the virtual function tables into the default data memory area, but this can be re-mapped as required. You can also direct the compiler to use a different section for C+ virtual function tables with the -section compiler switch. seg_sram This section is SDRAM memory. seg_heap In the default LDFs for ADSP-2136x/2137x/2147x/2148x processors, this section is the area from which memory allocation functions and operators (new, malloc(), and so on) allocate memory. In the default LDFs for ADSP-21020, ADSP-2106x, ADSP-2116x, and ADSP-2126x processors, the memory allocation functions and operators allocate memory from seg_dmda. In the VDK LDFs, this section is the area from which memory allocation functions and operators allocate memory. seg_flash In the LDFs for ADSP-213xx/ADSP-214xx processors, this section is flash memory.

2-17 VisualDSP++ 5.0 Linker and Utilities Manual Linker

seg_ext_swco In the LDFs for ADSP-214xx processors, this section is the external memory for sections that contain short-word instructions (using the variable instruction set). seg_ext_nwco In the LDFs for ADSP-214xx processors, this section is the external memory for sections that contain normal-word instructions (using the legacy instruction set). seg_ext_dmda In the LDFs for ADSP-214xx processors, this section is external memory used for global data qualified with the “dm” keyword. seg_ext_pmda In the LDFs for ADSP-214xx processors, this section is the external memory for global program data that is qualified with the “pm” keyword.

Other Memory Segments The compiler and libraries also use other data sections that are linked into one of the above memory segments. These data sections include: seg_ctdml The symbol “___ctor_NULL_marker” (located in the C++ run-time library) marks the end of the list of global and static constructors and is placed in this data section. The linker ensures that the contents of this data section are the last items in seg_ctdml. .gdt, .gdtl, .frt, .cht, and .edt These data sections are used to hold data used during the handling of exceptions. The linker places the contents of these data sections in seg_dmda. See “Blackfin Special “Table” Input Sections” on page 2-24.

VisualDSP++ 5.0 Linker and Utilities Manual 2-18 Link Target Description

Default Memory Segments for TigerSHARC Processors

Table 2-2 shows section mapping in the default .ldf file for a ADSP-TS101 processor (as a simplified example for TigerSHARC processors).

Table 2-2. Section Mapping in the Default TigerSHARC LDF File

Input Section Output Section Memory Segment

program code M0Code

data1 data1 M1Data

data2 data2 M2Data

mem_argv mem_argv M1Data

bsz bsz M1Data

bsz_init bsz_init M1Data

ctor data1 M1Data

ctor0 data1 M1Data

ctor1 data1 M1Data

ctor2 data1 M1Data

ctor3 data1 M1Data

ctor4 data1 M1Data

.gdt,.gdtl data1 M1Data

.frt data1 M1Data

.cht data1 M1Data

.edt data1 M1Data

.rtti data1 M1Data

vtbl vtbl M2DataA

Several input sections and memory segments are used in the default LDFs for ADSP-TSxxx processors must be present in user’s own LDFs. These sections are described in detail below.

2-19 VisualDSP++ 5.0 Linker and Utilities Manual Linker

For more information on stack and heap allocation, see “Allocation of memory for stacks and heaps in LDFs” in the C/C++ Compiler and Library Manual for TigerSHARC Processors. bsz This section is a BSS-style section for global zero-initialized data. bsz_init This section contains run-time initialization data. (See “-meminit” on page 2-61 for more information.) ctor This section contains the addresses of constructors that are called before the start of a C++ program (such as constructors for global and static objects). This section must be terminated with the symbol “___ctor_NULL_marker” (the default LDFs ensure this). It is required if compiling with C++ code.

When all ctor sections are merged, they form a table containing a list of all constructors for all global C++ objects. The table is used only at startup and can be placed in ROM. When linking, it is important that all ctor sections are merged in sequence (no other sections in between) and the run-time library or the VDK run-time library is placed with the first ctor section. Note that the default LDF’s “___ctor_NULL_marker” symbol is placed in a section named “ctorl” which must be the last of the ctor sections to be used as input. The final letter in this name is a lowercase “L”. data1 This section is the default location for global program data. data2 This section is the default location for global program data specified with the pm memory qualifier. mem_argv This section contains the command-line arguments that are used as part of profile-guided optimization (PGO).

VisualDSP++ 5.0 Linker and Utilities Manual 2-20 Link Target Description

program This section is the default location for program code. vtbl This section contains C++ virtual function tables. The default LDFs place the virtual function tables into the default data memory area but this can be re-mapped as required. You can also direct the compiler to use a different section for C+ virtual function tables, by using the -section compiler switch.

Other Memory Segments The compiler and libraries also use other data sections that are linked into one of the above memory segments. These data sections include: ctor1 This section contains the terminator for the ctor table section. It must be mapped immediately after the ctor sections. .gdt, .gdtl, .frt, .cht, .edt, and .rtti These data sections are used to hold data used during the handling of exceptions. The linker places the contents of these data sections in seg_dmda. See “Blackfin Special “Table” Input Sections” on page 2-24.

Default Memory Segments for Blackfin Processors

The default .ldf files in Blackfin/ldf show the mapping of input sections to output sections and memory segments. There are several input sections present in the default .ldf file and their uses are detailed below. See “Linker Description File” in Chapter 3, Linker Description File for more information on .ldf files and help on customization. Before customizing a default .ldf file, consider using the Expert Linker available from the VisualDSP++ IDDE. Generation and configuration of a custom .ldf file is available when creating a new project (or via the Project Options dialog box in VisualDSP++.

2-21 VisualDSP++ 5.0 Linker and Utilities Manual Linker

program This section is the default location for program code. data1 This section is the default location for global program data. cplb_code This section stores the run-time library’s cacheability protection lookaside buffer (CPLB) management routines. It is usually mapped into L1 Instruction SRAM. In particular, if CPLB replacement is possible, this section must be mapped to memory that is guaranteed to always be available; this means that it must be addressed by a locked CPLB. constdata This section is used for global data that is declared as constant and for lit- eral constants such as strings and array initializers. cplb_data This section stores CPLB configuration tables. In particular, the cplbtabx.doj files (where x indicates the target) mapped by .ldf files are placed into this section. L1_DATA_A This section is used to allow data to be mapped explicitly into L1 Data A SRAM using the SECTION directive. By default, the compiler does not generate data here. This section is analogous to L1_code. L1_DATA_B This section is similar to L1_DATA_A, except that it is used to map data into L1 Data B SRAM. voldata This section is used for data that may change due to external influences (such as DMA), and should not be placed into cached data areas. ctor This section contains addresses of C++ constructor functions which are to

VisualDSP++ 5.0 Linker and Utilities Manual 2-22 Link Target Description

be called before main() to construct static objects. The mapping of ctor must be followed directly by the mapping of ctor. bsz This section is used to map global zero-initialized data. This section does not actually contain data; it is zero-filled upon loading via the Visu- alDSP++ IDDE, via a command line, or when processed by the loader. bsz_init This section contains run-time initialization data. (See “-meminit” on page 2-61.) It is expected that this section is mapped into read-only memory. When a .dxe file has been processed by the Memory Initializer utility and the program starts running, other data sections (such as data1 and constdata) are initialized by data copied from this section. stack This section is the area where the run-time stack is located. Local variables, function parameters, and so on are stored here. heap This section is the area where the heap is located. Dynamically allocated data is placed here. noncache_code This section is mapped to areas of memory that cannot be cache and take program code. This section is used when you have a function that turns on the cache to ensure that the function itself does not reside in cache (as executing code from a cache memory address causes a hardware exception). sdram0 In most .ldf files and LDF configurations, this section allows code or data to be mapped explicitly into external memory by using the SECTION directive. This can be used to place large, infrequently used data or functions into external memory to free up valuable internal memory.

2-23 VisualDSP++ 5.0 Linker and Utilities Manual Linker

sdram0_bank{1|2|3} This section is used to map code and data into separate SDRAM banks which are defined when SDRAM is partitioned in the default .ldf files. sdram_bcz This section is the same as section bsz, except it is placed in SDRAM when SDRAM is enabled. sdram_shared This section is used to map code and data into the part of memory shared between core A and core B on multicore systems. vtbl This section contains C++ virtual function tables. The default .ldf files place the virtual function tables into the default data memory area but this can be re-mapped as required. You can also direct the compiler to use a different section for C+ virtual function tables by using the -section compiler switch.

Other Memory Segments The compiler and libraries also use other data sections that are linked into one of the above memory segments. These data sections include: ctorl This section contains the terminator for the ctor table section. It must be mapped immediately after the ctor section. .gdt, .gdtl, .frt, .cht, .edt, and .rtti These data sections are used to hold data used during the handling of exceptions. See “Blackfin Special “Table” Input Sections”.

Blackfin Special “Table” Input Sections The following “table” data sections are used to hold data used during the handling of exceptions. Generally, the linker maps these sections into read-only memory.

VisualDSP++ 5.0 Linker and Utilities Manual 2-24 Link Target Description

.gdt The .gdt section (global dispatch table) is used by the C++ exception library to determine which area of code to which a particular address belongs. This section must be contiguous in memory. .gdtl The .gdtl section contains the terminator for the .gdt table section. It must be mapped immediately after the .gdt section. .edt The .edt section (exception dispatch table) is used by the C++ exception library to map from try blocks to catch blocks. .cht The .cht section (catch handler types table) is used to map to the RTTI type information. The C++ exception library uses it to determine the types that correspond to catch entries for a try block. .frt The .frt section (function range table) is used by the C++ exception library during exception processing to unwind the stack of active functions. primio_atomic_lock The primio_atomic_lock section is used by the control variable that is used to ensure atomic file I/O. It must be in shared memory and not cached. mc_data The mc_data section is used to hold the core-specific storage on multi-core systems. .rtti The .rtti section is used by the C++ run-time type identification support, when enabled.

2-25 VisualDSP++ 5.0 Linker and Utilities Manual Linker

cplb The cplb section is in .ldf files for legacy reasons.

Input Sections in Blackfin Default LDFs for User Code/Data These sections are not normally used by the Blackfin compiler and L libraries. L1_data This section is used to allow global data to be mapped explicitly into L1 data SRAM using the section pragma or directive. This input section maps data to both banks A and B where present on the target. L1_data_a This section is not normally used by the compiler and libraries. L1_data_b This section is similar to L1_data_a, except that it is used to map data into L1 data B SRAM where it is present on the target chip. L1_code This section is used to allow code to be mapped explicitly into L1 code SRAM using the section pragma or directive. L1_bcz This section is used to map global zero initialized data into L1 data SRAM using the section pragma or directive. L2_bcz This section is used to map global zero-initialized data to L2 for parts which have L2 memory using the section pragma or directive. L2_sram This section can be used to map code and data into L2 for non-multicore parts that have L2 SRAM.

VisualDSP++ 5.0 Linker and Utilities Manual 2-26 Link Target Description

l2_sram This section can be used on a multicore system to map code and data into L2 for parts which have L2 memory. L2_sram_a This section is used to map code and data into the part of L2 memory reserved for core A on a multicore system. L2_sram_b This section is used to map code and data into the part of L2 memory reserved for code B on a multicore system. l2_shared This section is used to map code and data into the part of L2 memory shared between core A and core B.

Memory Characteristics Overview This section provides an overview of basic memory information (including addresses and ranges) for sample target architectures. Some portions of the processor memory are reserved. Refer to the L processor’s Hardware Reference for more information.

SHARC Memory Characteristics As an example of the SHARC memory architecture, the ADSP-21161 processor contains a large, dual-ported internal memory for single-cycle, simultaneous, independent accesses by the core processor and I/O processor. The dual-ported memory (in combination with three separate on-chip buses) allow two data transfers from the core and one transfer from the I/O processor in a single cycle. Using the I/O bus, the I/O processor provides data transfers between internal memory and the processor’s communication ports (link ports, serial ports, and external port) without hindering the processor core’s access to memory. The processor provides access to external memory through the processor’s external port.

2-27 VisualDSP++ 5.0 Linker and Utilities Manual Linker

The processor contains one megabit of on-chip SRAM, organized as two blocks of 0.5 Mbits. Each block can be configured for different combinations of code and data storage. All of the memory can be accessed as 16-, 32-, 48-, or 64-bit words. The memory can be configured in each block as a maximum of 16 Kwords of 32-bit data, 8 Kwords of 64-bit data, 32 Kwords of 16-bit data, 10.67 Kwords of 48-bit instructions (or 40-bit data), or combinations of different word sizes up to 0.5 Mbits. This gives a total for the complete internal memory: a maximum of 32 Kwords of 32-bit data, 16 Kwords of 64-bit data, 64 Kwords of 16-bit data, and 21 Kwords of 48-bit instructions (or 40-bit data). The processor features a 16-bit floating-point storage format that effectively doubles the amount of data that may be stored on-chip. A single instruction converts the format from 32-bit floating-point to 16-bit floating-point. While each memory block can store combinations of code and data, accesses are most efficient when one block stores data using the DM bus, (typically, Block 1) for transfers, and the other block (typically, Block 0) stores instructions and data using the PM bus. Using the DM bus and PM bus with one dedicated to each memory block assures single-cycle execution with two data transfers. In this case, the instruction must be available in the cache. Internal Memory ADSP-21161 processors have 2 Mbits of internal memory space; 1 Mbits are addressable. The 1 Mbits of memory is divided into two 0.5-Mbit blocks: Block 0 and Block 1. The additional 1 Mbits of the memory space is reserved on the ADSP-21161 processor. Table 2-3 shows the maximum

VisualDSP++ 5.0 Linker and Utilities Manual 2-28 Link Target Description

number of data or instruction words that can fit in each 0.5-Mbit internal memory block.

Table 2-3. Words Per 0.5-MBit Internal Memory Block

Word Type Bits Per Word Maximum Number of Words Per 0.5-Mbit Block Instruction 48-bits 10.67 Kwords Long word data 64-bits 8 Kwords Extended-precision normal word data 40-bits 10.67 Kwords Normal word data 32-bits 16 Kwords Short word data 16-bits 32 Kwords

External Memory Although the processor’s internal memory is divided into blocks, the processor’s external memory spaces are divided into banks. The internal memory blocks and the external memory spaces may be addressed by either data address generator (DAG). External memory banks are fixed sizes that can be configured for various waitstate and access configurations. The processor can address 254 Mwords of external memory space. External memory connects to the processor’s external port, which extends the processor’s 24-bit address and 32-bit data buses off the processor. The processor can make 8-, 16-, 32-, or 48-bit accesses to external memory for instructions and 8-, 16-, or 32-bit accesses for data. Table 2-4 shows the access types and words for processor’s external memory accesses. The processor’s DMA controller automatically packs external data into the appropriate word width during data transfer.

2-29 VisualDSP++ 5.0 Linker and Utilities Manual Linker

The external data bus can be expanded to 48 bits if the link ports L are disabled and the corresponding full-width instruction packing mode (IPACK) is enabled in the SYSCON register. Ensure that link ports are disabled when executing code from external 48-bit memory.

Table 2-4. Internal-to-External Memory Word Transfers

Word Type Transfer Type Packed instruction 32-, 16-, or 8-to-48 bit packing Normal word data 32-bit word in 32-bit transfer Short word data Not supported

The total addressable space for the fixed external memory bank sizes depends on whether SDRAM or non-SDRAM (such as SRAM, SBSRAM) is used. Each external memory bank for SDRAM can address 64 Mwords. For non-SDRAM memory, each bank can address up to 16 Mwords. The remaining 48 Mwords are reserved. These reserved addresses for non-SDRAM accesses are aliased to the first 16 Mspaces within the bank.

TigerSHARC Memory Characteristics As an example of the TigerSHARC memory architecture, the ADSP-TS101 processor has three internal memory blocks: M0, M1, and M2. Each memory block consists of 2 Mbits of memory space and is configured as 64 Kwords (each 32 bits in width). There are three separate internal 128-bit data buses, each connected to one of the memory blocks. Memory blocks can store instructions and data interchangeably, with one access per memory block per cycle. If the programmer ensures that program and data are in different memory blocks, data access can occur at the same time as program fetch. Therefore, in one cycle, up to three 128-bit transfers can occur within the core (two data transfers and one program instruction transfer).

VisualDSP++ 5.0 Linker and Utilities Manual 2-30 Link Target Description

The I/O Processor can use only one internal bus at a time, and the I/O Processor competes with the core for use of the internal bus. Therefore, in one cycle, the processor can fetch four 32-bit instructions and load or store 256 bits of data (four 64-bit words, eight 32-bit words, sixteen 16-bit words, or thirty-two 8-bit words). The TigerSHARC processor 32-bit address bus provides an address space of four gigawords. This address space is common to a cluster of Tiger- SHARC processors that share the same cluster bus. The zones in the memory space are made up of the following regions. • External memory bank space—the region for standard addressing of off-chip memory (including SDRAM, MB0, MB1, and host) • External multiprocessor space—the on-chip memory of all other TigerSHARC processors connected in a multiprocessor system • Internal address space—the region for standard internal addressing In the example system, the ADSP-TS101 processor has internal memory addresses from 0x0 to 0x17FFFF. Refer to Table 2-5.

Table 2-5. ADSP-TS101 Processor Memory Structure

Block Range Word Size

M0 memory block 0x0000 0000 - 0x0000 FFFF 32-bit instructions

0x0001 0000 - 0x0007 FFFF Reserved

M1 memory block 0x0008 0000 - 0x0008 FFFF 32-bit instructions

0x0009 0000 - 0x0009 FFFF Reserved

M2 memory block 0x0010 0000 - 0x0010 FFFF 32-bit instructions

0x0011 0000 - 0x0017 FFFF Reserved

Internal registers 0x0018 0000 - 0x0018 07FF Control, status, and I/O registers. This cannot be used in LDFs. Internal registers are memory accessible in MP space only.

0x0018 0800 - 0x01BF FFFF Reserved

2-31 VisualDSP++ 5.0 Linker and Utilities Manual Linker

Table 2-5. ADSP-TS101 Processor Memory Structure (Cont’d)

Block Range Word Size

0x01C0 0000 - 0x03FF FFFF Broadcast and multiprocessor (not used in LDF)

SDRAM 0x0400 0000 - 0x07FF FFFF 32-bit instructions

Blackfin Memory Characteristics Details of the Blackfin processor memory characteristics can be found in the data sheets for individual processors, available in the appropriate Hardware Reference or at: www.analog.com/processors/productsDatasheets/dataSheets.html.

Linker MEMORY{} Command in an LDF Referring to information in sections “Memory Usage and Default Mem- ory Segments” and “Memory Characteristics Overview”, you can specify the target’s memory with the MEMORY{} command for any of target processor architectures (Listing 2-1, Listing 2-2, and Listing 2-3 provide code examples for specific processors).

Listing 2-1. ADSP-21161 MEMORY{} Command Code

MEMORY { seg_rth { TYPE(PM RAM) START(0x00040000) END(0x000400ff) WIDTH(48) } seg_init { TYPE(PM RAM) START(0x00040100) END(0x000401ff) WIDTH(48) } seg_int_code { TYPE(PM RAM) START(0x00040200) END(0x00040287) WIDTH(48) } seg_pmco { TYPE(PM RAM) START(0x00040288) END(0x000419ff) WIDTH(48) }

VisualDSP++ 5.0 Linker and Utilities Manual 2-32 Link Target Description

seg_pmda { TYPE(PM RAM) START(0x00042700) END(0x00043fff) WIDTH(32) } seg_dmda { TYPE(DM RAM) START(0x00050000) END(0x00051fff) WIDTH(32) } seg_heap { TYPE(DM RAM) START(0x00052000) END(0x00052fff) WIDTH(32) } }

Listing 2-2. ADSP-TS101 MEMORY{} Command

MEMORY { /* Internal memory blocks are 0x10000 (64K bytes) */ /* Start of TS101_memory.ldf */

M0Code {TYPE(RAM) START(0x00000000) END(0x0000FFFF) WIDTH(32)} M1Data {TYPE(RAM) START(0x00080000) END(0x0008BFFF) WIDTH(32)} M1Stack {TYPE(RAM) START(0x0008C000) END(0x0008FFFF) WIDTH(32)} M2Data {TYPE(RAM) START(0x00100000) END(0x0010BFFF) WIDTH(32)} M2Heap {TYPE(RAM) START(0x0010C000) END(0x0010C7FF) WIDTH(32)} M2Stack {TYPE(RAM) START(0x0010C800) END(0x0010FFFF) WIDTH(32)} SDRAM {TYPE(RAM) START(0x04000000) END(0x07FFFFFF) WIDTH(32)} MS0 {TYPE(RAM) START(0x08000000) END(0x0BFFFFFF) WIDTH(32)} MS1 {TYPE(RAM) START(0x0C000000) END(0x0FFFFFFF) WIDTH(32)}

/* end of TS101_memory.ldf file */ }

Listing 2-3. ADSP-BF533 MEMORY{} Command Code

MEMORY /* Define/label system memory */ { /* List of global Memory Segments */ MEM_L2_CODE { TYPE(RAM) START(0xF0000000) END(0xF002FFFF) WIDTH(8) }

2-33 VisualDSP++ 5.0 Linker and Utilities Manual Linker

MEM_L1_DATA_A { TYPE(RAM) START(0xFF800000) END(0xFF803FFF) WIDTH(8) } MEM_L1_DATA_B { TYPE(RAM) START(0xFF900000) END(0xFF903FFF) WIDTH(8) } MEM_HEAP { TYPE(RAM) START(0xF0030000) END(0xF0037FFF) WIDTH(8) } MEM_STACK { TYPE(RAM) START(0xF0038000) END(0xF003DFFF) WIDTH(8) } MEM_ARGV { TYPE(RAM) START(0xF003FE00) END(0xF003FFFF) WIDTH(8) } MEM_SDRAM0 { TYPE(RAM) START(0x00000004) END(0x07FFFFFF) WIDTH(8) } }

The above examples apply to the preceding discussion of how to L write a MEMORY{} command and to the following discussion of the SECTIONS{} command. The SECTIONS{} command is not atomic; it can be interspersed with other directives, including location counter information. You can define new symbols within the .ldf file. These examples define the starting stack address, the highest possible stack address, and the heap’s starting location and size. These newly-created symbols are entered in the executable’s symbol table.

Entry Address In releases prior to VisualDSP++ 4.5, the entry address was filled in from a global symbol “start” (no underscore), if present. The “start” symbol could be a global file symbol or an LDF symbol.

VisualDSP++ 5.0 Linker and Utilities Manual 2-34 Link Target Description

Currently, the entry address field can also be set using:

• The -entry command-line switch (on page 2-58), where option’s argument is a symbol.

• The ENTRY(symbol) command (on page 3-40) in the .ldf file. If -entry and ENTRY() are both present, they must be the same. Neither overrides the other. If there is a mismatch, the linker detects an error.

• In the absence of the -entry switch or the ENTRY() command, the value of the global file symbol start, or LDF symbol start, is used, if present. • If none of the above is used, the address is 0. Multiprocessor/Multicore Applications The -entry switch for a multiprocessor/multi-core .ldf file applies the same entry address to all processors. If the entry addresses differ (multiprocessor systems), use ENTRY() commands in the .ldf file – do not use the -entry switch.

If the -entry switch is specified, it is an error if any of the processors uti- lize an ENTRY() command with a different specification.

Wildcard Characters The linker supports the use of wildcards in input section name specifica- tions in the .ldf file. The * and ? wildcard characters are provided on input section names so that you can specify multiple input sections.

* – Matches any number of characters

? – Matches any one character For information about wildcard characters used (and an example) with the INPUT_SECTIONS command, see “INPUT_SECTIONS()” on page 3-64.

2-35 VisualDSP++ 5.0 Linker and Utilities Manual Linker

Placing Code on the Target

Use the SECTIONS{} command to map code and data to the physical memory of a processor in a processor system.

To write a SECTIONS{} command: 1. List all input sections defined in the source files.

• Assembly files – List each assembly code .SECTION directive, identify its memory type (PM or CODE, or DM or DATA), and note when location is critical to its operation. These .SEC- TIONS portions include interrupt tables, data buffers, and on-chip code or data. (See “Specifying Two Buffers in Dif- ferent Memory Segments” on page 2-41 for TigerSHARC-specific information.) • C/C++ source files – The compiler generates sections with the name “program” or “code” for code, and the names “data1” and “data2” for data. These sections correspond to your source when you do not specify a section by means of the optional section() extension. 2. Compare the input sections list to the memory segments specified in the MEMORY{} command. Identify the memory segment into which each .SECTION must be placed. 3. Combine the information from these two lists to write one or more SECTIONS{} commands in the .ldf file.

SECTIONS{} commands must appear within the context of L the PROCESSOR{} or SHARED_MEMORY() command.

Listing 2-4 presents a SECTIONS{} command that would work with the MEMORY{} command in Listing 2-1.

VisualDSP++ 5.0 Linker and Utilities Manual 2-36 Link Target Description

Listing 2-4. ADSP-21161 SECTIONS{} Command in the LDF

SECTIONS { /* Begin output sections */ seg_rth { // run-time header and interrupt table INPUT_SECTIONS( $OBJS(seg_rth) $LIBS(seg_rth)) } >seg_rth seg_init { // Initialization ldf_seginit_space = . ; INPUT_SECTIONS( $OBJS(seg_init) $LIBS(seg_init)) } >seg_init seg_init_code { // Initialization data INPUT_SECTIONS( $OBJS(seg_init_code) $LIBS(seg_init_code)) } >seg_init_code seg_pmco { // PM code INPUT_SECTIONS( $OBJS(seg_pmco) $LIBS(seg_pmco)) } >seg_pmco seg_pmda { // PM data INPUT_SECTIONS( $OBJS(seg_pmda) $LIBS(seg_pmda)) } >seg_pmda .bss ZERO_INIT { INPUT_SECTIONS( $OBJS(.bss) $LIBS(.bss)) } >seg_dmda seg_dmda { // DM data INPUT_SECTIONS( $OBJS(seg_dmda) $LIBS(seg_dmda)) } >seg_dmda

heap { // allocate a heap for the application ldf_heap_space = .; ldf_heap_length = MEMORY_SIZEOF(seg_heap); ldf_heap_end = ldf_heap_space + ldf_heap_length - 1;

2-37 VisualDSP++ 5.0 Linker and Utilities Manual Linker

} > seg_heap; } // end sections

Listing 2-5 presents a SECTIONS{} command that would work with the MEMORY{} command in Listing 2-2.

Listing 2-5. ADSP-TS101 SECTIONS{} Command in the LDF

SECTIONS { /* List of sections for processor P0 */ sec_rth {INPUT_SECTIONS ( $OBJECTS(rth))} > seg_rth sec_code {INPUT_SECTIONS ( $OBJECTS(code)} > seg_code sec_code2 {INPUT_SECTIONS ( $OBJECTS(y_input)} > seg_code sec_data1 {INPUT_SECTIONS ( $OBJECTS(data1))} > seg_data1 } }

Listing 2-6 presents a SECTIONS{} command that would work with the MEMORY{} command in Listing 2-3.

Listing 2-6. ADSP-BF535 SECTIONS{} Command in the LDF

SECTIONS { /* List of sections for processor P0 */

L1_code { INPUT_SECTION_ALIGN(2) /* Align all code sections on 2 byte boundary */ INPUT_SECTIONS( $OBJECTS(program) $LIBRARIES(program)) INPUT_SECTION_ALIGN(1) INPUT_SECTIONS( $OBJECTS(data1) $LIBRARIES(data1)) INPUT_SECTION_ALIGN(1) INPUT_SECTIONS( $OBJECTS(constdata) $LIBRARIES(constdata))

VisualDSP++ 5.0 Linker and Utilities Manual 2-38 Link Target Description

INPUT_SECTION_ALIGN(1) INPUT_SECTIONS( $OBJECTS(ctor) $LIBRARIES(ctor) ) } >MEM_L2_CODE

program { // Align all code sections on 2 byte boundary INPUT_SECTION_ALIGN(4) INPUT_SECTIONS( $OBJECTS(program) $LIBRARIES(program)) INPUT_SECTIONS( $OBJECTS(data1) $LIBRARIES(data1)) INPUT_SECTIONS( $OBJECTS(cplb) $LIBRARIES(cplb)) INPUT_SECTIONS( $OBJECTS(cplb_code) $LIBRARIES(cplb_code)) INPUT_SECTIONS( $OBJECTS(cplb_data) $LIBRARIES(cplb_data)) INPUT_SECTIONS( $OBJECTS(constdata) $LIBRARIES(constdata)) INPUT_SECTIONS( $OBJECTS(voldata) $LIBRARIES(voldata)) } >MEM_PROGRAM

stack { ldf_stack_space = .; ldf_stack_end = ldf_stack_space + MEMORY_SIZEOF(MEM_STACK) - 4; } >MEM_STACK

heap { /* Allocate a heap for the application */ ldf_heap_space = .; ldf_heap_end = ldf_heap_space + MEMORY_SIZEOF(MEM_HEAP) - 1; ldf_heap_length = ldf_heap_end - ldf_heap_space; } >MEM_HEAP

2-39 VisualDSP++ 5.0 Linker and Utilities Manual Linker

argv { /* Allocate argv space for the application */ ldf_argv_space = .; ldf_argv_end = ldf_argv_space + MEMORY_SIZEOF(MEM_ARGV) - 1; ldf_argv_length = ldf_argv_end - ldf_argv_space; } >MEM_ARGV

} /* end SECTIONS */

VisualDSP++ 5.0 Linker and Utilities Manual 2-40 Link Target Description

Specifying Two Buffers in Different Memory Segments On TigerSHARC processors, the linker supports optimized memory placement, using the .SEPARATE_MEM_SEGMENTS assembler directive.

• The .SEPARATE_MEM_SEGMENTS assembler directive (or the compiler pragma #pragma separate_mem_segments) specifies two buffers directing the linker to place the buffers into different memory segments. For example,

.SECTION data1;

.VAR buf1;

.VAR buf2;

.EXTERN buf3;

.SEPARATE_MEM_SEGMENTS(buf1, buf2);

.SEPARATE_MEM_SEGMENTS(buf1, buf3); • The set of available memory segments for each buffer is defined by using the linker’s “one-to-many” feature—mapping input section(s) that contain the buffer into multiple memory segments. For example,

data2 {

INPUT_SECTIONS( $OBJECTS(data1) )

} >M2DataA

data4 {

INPUT_SECTIONS( $OBJECTS(data1) )

} >M4DataA

2-41 VisualDSP++ 5.0 Linker and Utilities Manual Linker

• The linker tries to satisfy placement constraint requirements by allocating the buffers to different memory segments. 1. If the linker fails to satisfy any or all of the requirements, the linker produces a warning.

2. All symbols mentioned in .SEPARATE_MEM_SEGMENTS are mapped before anything else by the linker (with the exception of absolute placement).

3. Reference to symbol in .SEPARATE_MEM_SEGMENTS is a weak reference. If such symbol is defined in a library, the linker does NOT bring the symbol from the library (unless the symbol is referenced directly or indirectly from an object file. 4. The linker ignores the cases where the symbol is mentioned in the .SEPARATE_MEM_SEGMENTS assembler directive is undefined or is not defined in an input section (for example, as an LDF symbol). See “Pragmas” in Chapter 1 of the VisualDSP++ 5.0 C/C++ Com- L piler and Library Manual for TigerSHARC Processors for more information.

Linking with Attributes – Overview

Attributes are used within the .ldf file to create virtual subsets from the usual input sources. Attributes are associated with .doj files, including those within the library. Once created, these subsets exist for the duration of the link and can be used anywhere a library or object list normally appears within an .ldf file.

Attributes are used within the .ldf file to reduce the usual set of input files into more manageable subsets. Inputs are in two forms (objects and libraries) both of which appear in lists within the .ldf file. Filters can be applied to these lists to winnow out momentarily-undesirable objects.

VisualDSP++ 5.0 Linker and Utilities Manual 2-42 Link Target Description

An attribute is a name/value pair of strings. A valid attribute name is a valid C identifier. Attribute names and attribute values are case-sensitive. Windows filenames can be used as values, with care and consistency.

An attribute is associated with an object (.doj), but not with a library (.dlb), not with a symbol name, and not with an ELF section. An object has zero or more attributes associated with it. A given object may have more than one attribute with the same name associated with it. Using attributes, the filtering process can be used to remove some objects from consideration, providing that the same objects are not included else- where via other filters (or through unfiltered mappings). A filter operation is done with curly braces, and can be used to define sub-lists and sub-libraries. It may also be used in INPUT_SECTIONS commands (refer to “INPUT_SECTIONS()” on page 3-64).

The linker reads the .ldf file and uses the {...} filter commands (for example, INPUT_SECTIONS commands) to eliminate some input objects from consideration before resolving symbols. The linker does not change its behavior if no filter commands are present in the .ldf file.

Profile-Guided Optimization Support

The SHARC, TigerSHARC, and Blackfin processor architectures support profile-guided optimization (PGO). PGO is the process of gathering information about a running application over many invocations of the executable with different input data, and then re-optimizing it using the gathered information. The process relies upon the same application being run with different data sets, which often means that the application acts upon sample data sets stored in files. More specifically, it means that the application is instructed to process each file via command-line options passed to main().

2-43 VisualDSP++ 5.0 Linker and Utilities Manual Linker

The .ldf files and the VisualDSP++ IDDE collaborate to provide support for command-line arguments. Under normal circumstances, a typical embedded program is not interested in command-line arguments, and receives none. In these normal cases, the run-time header invokes a function to parse a global string __argv_string[] and finds it empty.

To support PGO, the LDF option IDDE_ARGS can be used to define a memory segment called MEM_ARGV, and __argv_string[] is mapped directly to the start of this section. The VisualDSP++ IDDE follows the convention that command-line arguments can be passed to an application by writing the argument string into memory starting at the beginning of MEM_ARGV. For more information on profile-guided optimization, refer to the L VisualDSP++ 5.0 C/C++ Compiler and Library Manual for the appropriate processor architecture.

Passing Arguments for Simulation or Emulation

The symbol _argv_string is a null-terminated string that, if it contains anything other than null, will be split at each space character and placed in the argv[] array that gets passed to the main function on system startup.

Linker Command-Line Reference

This section provides reference information, including: • “Linker Command-Line Syntax” on page 2-45 • “Linker Command-Line Switches” on page 2-49

VisualDSP++ 5.0 Linker and Utilities Manual 2-44 Linker Command-Line Reference

When you use the linker via the VisualDSP++ IDDE, the settings L on the Link page of the Project Options dialog box correspond to linker command-line switches. Provided here is the detailed descriptions of the linker’s command-line switches and their syntax. For more information, refer to VisualDSP++ online Help.

Linker Command-Line Syntax

Run the linker by using one of the following normalized formats of the linker command line.

linker -proc processor -switch [-switch …] object [object …] linker -T target.ldf -switch [-switch …] object [object …]

The linker command requires -proc processor or a -T L to proceed. If the command line does not include -proc processor, the .ldf file following the -T switch must contain a -ARCHITECTURE command. The linker command may contain both, but then the ARCHITECTURE() command in the .ldf file must match the -proc processor.

Use -proc processor instead of the deprecated -Darchitecture switch on the command line to select the target processor. See Table 2-7 on page 2-51 for more information. All other switches are optional, and some commands are mutually exclusive. The following are example linker commands. linker -proc ADSP-21161 p0.doj -T target.ldf -t -o program.dxe linker -proc ADSP-TS201 p0.doj -T target.ldf -t -o program.dxe linker -proc ADSP-BF535 p0.doj -T target.ldf -t -o program.dxe

The linker command line (except for file names) is case sensitive. L For example, linker -t differs from linker -T.

2-45 VisualDSP++ 5.0 Linker and Utilities Manual Linker

The linker can be controlled by the compiler via the -flags-link command-line switch, which passes explicit options to the linker. For more information, refer to Chapter 1 of the VisualDSP++ 5.0 C/C++ Compiler and Library Manual for the appropriate processor. When using the linker’s command line, be familiar with the following topics: • “Command-Line Object Files” • “Command-Line File Names” • “Object File Types” on page 2-49

Command-Line Object Files The command line must identify at least one (typically more) object file(s) to be linked together. These files may be of several different types.

• Standard object (.doj) files produced by the assembler

• One or more libraries (archives), each with a .dlb extension. Examples include the C run-time libraries and math libraries included with VisualDSP++. You may create libraries of common or specialized objects. Special libraries are available from DSP algorithm vendors. For more information, see Chapter 6, “Archiver”.

• An executable (.dxe) file to be linked against. Refer to $COMMAND_LINE_LINK_AGAINST in “Built-In LDF Macros” on page 3-30. Object File Names

VisualDSP++ 5.0 Linker and Utilities Manual 2-46 Linker Command-Line Reference

An object file name may include: • The drive, directory path, file name, and file extension • The directory path may be an absolute path or a path relative to the directory from which the linker is invoked • Long file names enclosed within straight quotes If the file exists before the link begins, the linker opens the file to verify its type before processing the file. Table 2-6 lists valid file extensions used by the linker.

Command-Line File Names Some linker switches take a file name as a parameter. Table 2-6 lists the types of files, names, and extensions that the linker expects on file name arguments. The linker follows the conventions for file extensions in Table 2-6.

Table 2-6. File Extension Conventions

Extension File Description

.dlb Library (archive) file

.doj Object file

.dxe Executable file

.ldf Linker Description File

.ovl Overlay file

.sm Shared memory file

2-47 VisualDSP++ 5.0 Linker and Utilities Manual Linker

The linker supports relative and absolute directory names, default directories, and user-selected directories for file search paths. File searches occur in the following order. 1. Specified path – If the command line includes relative or absolute path information, the linker searches that location for the file. 2. Specified directories – If you do not include path information on the command line and the file is not in the default directory, the linker searches for the file in the search directories specified with the -L (path) command-line switch, and then searches directories specified by SEARCH_DIR commands in the .ldf file. Directories are searched in order of appearance on the command line or in the .ldf file. 3. Default directory – If you do not include path information in the .ldf file named by the -T switch, the linker searches for the .ldf file in the current working directory. If you use a default .ldf file (by omitting LDF information in the command line and instead specifying -proc ), the linker searches in the processor-specific LDF directory; for example, $ADI_DSP/Blackfin/ldf. For more information on file searches, see “Built-In LDF Macros” on page 3-30. When providing input or output file names as command-line parameters: • Use a space to delimit file names in a list of input files. • Enclose file names that contain spaces within straight quotes; for example, "long file name". • Include the appropriate extension to each file. The linker opens existing files and verifies their type before processing. When the linker creates a file, it uses the file extension to determine the type of file to create.

VisualDSP++ 5.0 Linker and Utilities Manual 2-48 Linker Command-Line Reference

Object File Types The linker handles an object (file) by its file type. File type is determined by the following rules. • Existing files are opened and examined to determine their type. Their names can be anything. • Files created during the link are named with an appropriate extension and are formatted accordingly. A map file is generated in XML format only and is given an .xml extension. An executable is written in the ELF format and is given a .dxe extension.

The linker treats object (.doj) files and library (.dlb) files that appear on the command line as object files to be linked. The linker treats executable (.dxe) files and shared memory (.sm) files on the command line as executables to be linked against.

For more information on objects, see the $COMMAND_LINE_OBJECTS macro. For information on executables, see the $COMMAND_LINE_LINK_AGAINST macro. Both are described in “Built-In LDF Macros” on page 3-30.

If link objects are not specified on the command line or in the .ldf file, the linker generates appropriate informational or error messages.

Linker Command-Line Switches

This section describes the linker’s command-line switches. Table 2-7 on page 2-51 briefly describes each switch with regard to case sensitivity, equivalent switches, switches overridden or contradicted by the one described, and naming and spacing constraints for parameters. The linker provides switches to select operations and modes. The standard switch syntax is -switch [argument].

2-49 VisualDSP++ 5.0 Linker and Utilities Manual Linker

Rules: • Switches may be used in any order on the command line. Items in brackets [] are optional. Items in italics are user-definable and are described with each switch. • Path names can be relative or absolute. • File names containing white space or colons must be enclosed by double quotation marks, though relative path names such as ../../test.dxe do not require double quotation marks. Different switches require (or prohibit) white space between the L switch and its parameter. Example: linker -proc ADSP-BF535 p0.doj p1.doj p2.doj -T target.ldf -t -o program.dxe

Note the difference between the -T and the -t switches. The command calls the linker as follows:

• -proc ADSP-BF535 Specifies the processor

• p0.doj, p1.doj, and p2.doj Links three object files into an executable file

• -T target.ldf Uses a secondary LDF to specify executable program placement

• -t Turns on trace information, echoing each link object’s name to stdout as it is processed

• -o program.dxe Specifies the name of the linked executable file

VisualDSP++ 5.0 Linker and Utilities Manual 2-50 Linker Command-Line Reference

Typing linker without any switches displays a summary of command-line options. Using no switches is the same as typing linker -help.

Linker Switch Summary and Descriptions Table 2-7 briefly describes each linker switch. Each individual switch is described in detail following this table. See “Project Builds” on page 2-7 for information on the VisualDSP++ Project Options dialog box.

Table 2-7. Linker Command-Line Switch Summary

Switch Description More Info

@file Uses the specified file as input on the command line on page 2-53

-DprocessorID Specifies the target processor ID. The use of on page 2-53 -proc processorID is recommended.

-L path Adds the path name to search libraries for objects on page 2-54

-M Produces dependencies on page 2-54

-MM Builds and produces dependencies on page 2-54

-Map file Outputs a map of link symbol information to a file on page 2-55

-MDmacro[=def] Defines and assigns value def to a preprocessor on page 2-55 macro

-MUDmacro Undefines the preprocessor macro on page 2-56

-S Omits debugging symbols from the output file on page 2-56

-T filename Identifies the LDF to be used on page 2-56

-Werror number Promotes the specified warning message to an error on page 2-57

-Wwarn number Demotes the specified error message to a warning on page 2-57

-Wnumber Selectively disables warnings by one or more mes- on page 2-57 sage numbers. For example,-W1010 disables warning message li1010.

-e Eliminates unused symbols from the executable on page 2-57

-ek secName Specifies a section name in which elimination on page 2-57 should not take place

2-51 VisualDSP++ 5.0 Linker and Utilities Manual Linker

Table 2-7. Linker Command-Line Switch Summary (Cont’d)

Switch Description More Info

-es secName Names input sections (secName list) to which the on page 2-58 elimination algorithm is applied

-ev Eliminates unused symbols verbosely on page 2-58

-entry Specifies entry address where an argument can be on page 2-58 either a symbol or an address

-flag-meminit Passes each comma-separated option to the Memory on page 2-59 Initializer utility

-flag-pp Passes each comma-separated option to the prepro- on page 2-59 cessor

-h Outputs the list of command-line switches and exits on page 2-59 -help

-i path Includes search directory for preprocessor include on page 2-59 files

-ip Fills fragmented memory with individual data on page 2-59 objects that fit

-jcs2l Converts out-of-range short calls and jumps to the on page 2-60 longer form. It also allows the linker to convert out-of-range branches to indirect calls and jump sequences.

-jcs2l+ Same as -jcs2l on page 2-60

-keep symName Keeps symbols from being eliminated on page 2-60

-meminit Causes post-processing of the executable file on page 2-61

-nomemcheck Turns off LDF memory checking on page 2-61

-o filename Outputs the named executable file on page 2-61

-od filename Specifies the output directory on page 2-61

-pp Stops after preprocessing on page 2-62

-proc processor Selects a target processor on page 2-62

-reserve-null Directs the linker to reserve 4 addressable units on page 2-62 (words) in memory at address 0x0

-s Strips symbol information from the output file on page 2-62

VisualDSP++ 5.0 Linker and Utilities Manual 2-52 Linker Command-Line Reference

Table 2-7. Linker Command-Line Switch Summary (Cont’d)

Switch Description More Info

-save-temps Saves temporary output files on page 2-63

-si-revision version Specifies silicon revision of the specified processor on page 2-63

-sp Skips preprocessing on page 2-64

-t Outputs the names of link objects on page 2-64

-tx Outputs full names of link objects on page 2-64

-v Verbose: Outputs status information on page 2-64 -verbose

-version Outputs version information and exits on page 2-64

-warnonce Warns only once for each undefined symbol on page 2-64

-xref Produces a cross-reference file on page 2-65

The following sections provide the detailed descriptions of the linker’s command-line switches.

@filename

The @ switch causes the linker to treat the contents of filename as input to the linker command line. The @ switch circumvents environmental command-line length restrictions. The filename may not start with “linker” (that is, it cannot be a linker command line). White space (including “newline”) in filename serves to separate tokens.

-Dprocessor

The -Dprocessor (define processor) switch specifies the target processor (architecture); for example, -DADSP-BF533.

The -proc processor switch (on page 2-62) is a preferred option L to be used as a replacement for the -Dprocessor command-line entry to specify the target processor.

2-53 VisualDSP++ 5.0 Linker and Utilities Manual Linker

White space is not permitted between -D and processor. The architecture entry is case sensitive and must be available in your VisualDSP++ installation. This switch (or -proc processor switch) must be used if no .ldf file is specified on the command line. (See -T on page 2-56.) This switch (or -proc processor switch) must be used if the specified .ldf file does not specify ARCHITECTURE(). Architectural inconsistency between this switch and the .ldf file causes an error.

-L path

The -L path (search directory) switch adds a path name to search libraries and objects. This switch is case-sensitive and spacing is unimportant. The path parameter enables searching for any file, including the .ldf file itself. To add multiple search paths, repeat the switch or specify a list of paths terminated by semicolons (;) with the final semicolon being optional. The paths named with this switch are searched before arguments in the SEARCH_DIR{} command.

-M

The -M (generate make rule only) switch directs the linker to check a dependency and to output the result to stdout.

-MM

The -MM (generate make rule and build) switch directs the linker to output a rule, which is suitable for the make utility, describing the dependencies of the source file. The linker checks for a dependency, outputs the result to stdout, and performs the build. The only difference between -MM and -M actions is that the linking continues with -MM. See “-M” for more information.

VisualDSP++ 5.0 Linker and Utilities Manual 2-54 Linker Command-Line Reference

-Map filename

The -Map filename (generate a memory map) switch directs the linker to output a memory map of all symbols. The map file name corresponds to the filename argument. The linker generates the map file in XML format only. For example, if the file name argument is test, the map file name is test.map.xml.

Opening an .xml map file in a Web browser provides an organized view of the map file. By using hyperlinks, it becomes easy to quickly find any relevant information. Since the format of .xml files can be extended between VisualDSP++ releases, the map file is dependant on particular installation of VisualDSP++. Thus, the .xml map file can be used only on the machine on which it was generated. In order to view the map file on a different machine, the file should be transformed to HTML format using the xmlmap2html.exe command-line utility. The utility makes it possible to view the map on virtually any machine with any browser. XSLT is a language for transforming XML documents. VisualDSP++ includes the following XSLT files for transforming and displaying the XML map files produced by the linker in a browser.

• System/linker_map_ss1.xsl Does not display symbols that start with a dot. This file is the default.

• /System/linker_map_ss2.xsl Cause all symbols to be displayed. Note that the compiler and libraries may use such symbols for local data and code.

-MDmacro[=def]

The -MDmacro[=def] (define macro) switch declares and assigns value def to the preprocessor macro named macro. For example, -MDTEST=BAR exe-

2-55 VisualDSP++ 5.0 Linker and Utilities Manual Linker

cutes the code following #ifdef TEST==BAR in the LDF (but not the code following #ifdef TEST==XXX).

If =def is not included, macro is declared and set to “1” to ensure the code following #ifdef TEST is executed. This switch may be repeated.

-MUDmacro

The -MUDmacro (undefine macro) switch undefines the preprocessor macro where macro specifies a name. For example, -MUDTEST undefines macro TEST. The switch is processed after all -MDmacro switches have been processed. The -MUDmacro switch may be repeated on the command line.

-S

The -S (strip debug symbol) switch directs the linker to omit source debugging information from the output file. Compare this switch to the -s switch on page 2-62.

-T filename

The -T filename (linker description file) switch directs the linker to use filename as the name of the .ldf file. The .ldf file specified following the -T switch must contain an ARCHITECTURE() command if the command line does not have -proc . The linker requires the -T switch when linking for a processor for which no VisualDSP++ support has been installed. In such cases, the processor ID does not appear in the Target processor field of the Project Options dialog box.

The filename must exist and be found (for example, via the -L option). White space must appear before filename. A file’s name is unconstrained, but must be valid. For example, a.b works if it is a valid .ldf file, where .ldf is a valid extension but not a requirement.

VisualDSP++ 5.0 Linker and Utilities Manual 2-56 Linker Command-Line Reference

-Werror [number]

The -Werror switch directs the linker to promote the specified warning message to an error. The number argument specifies the message to promote.

-Wwarn [number]

The -Wwarn switch directs the linker to demote the specified error message to a warning. The number argument specifies the message to demote.

-Wnumber[,number]

The -Wnumber or -wnumber (warning suppression) switches selectively disables warnings specified by one or more message numbers. For example, -W1010 disables warning message li1010. Optionally, this switch accepts a list, such as [,number ...].

-e

The -e switch directs the linker to eliminate unused symbols from the executable file. In order for the C and C++ run-time libraries to work properly, L the following symbols should be retained with the “KEEP()” LDF command (described on page 3-42): ___ctor_NULL_marker and ___lib_end_of_heap_descriptions.

-ek sectionName

The -ek sectionName (no elimination) switch specifies a section to which the elimination algorithm is not applied. Both this switch and the KEEP_SECTIONS() LDF command (on page 3-42) may be used to specify a section name in which elimination should not take place.

2-57 VisualDSP++ 5.0 Linker and Utilities Manual Linker

-es sectionName

The -es sectionName (eliminate listed section) switch specifies a section to which the elimination algorithm is to be applied. This switch restricts elimination to the named input sections. The -es switch may be used on a command line more than once. In the absence of the -es switch or the ELIMINATE_SECTIONS() LDF command (on page 3-40), the linker applies elimination to all sections. Both this switch and the ELIMINATE_SECTIONS() LDF command may be used to specify sections from which unreferenced code and data are to be eliminated. In order for the C and C++ run-time libraries to work properly, L the following symbols should be retained with the “KEEP()” LDF command (described on page 3-42): ___ctor_NULL_marker and ___lib_end_of_heap_descriptions

-entry

The -entry switch indicates the entry address where an argument can be either a symbol or an address.

-ev

The -ev switch directs the linker to eliminate unused symbols and reports on each eliminated symbol.

-flags-meminit -opt1[,-opt2...]

The -flags-meminit switch passes each comma-separated option to the Memory Initializer utility. (For more information, see “Memory Initial- izer” in Chapter 7, Memory Initializer.)

-flags-pp-opt1[,-opt2...]

The -flags-pp switch passes each comma-separated option to the preprocessor.

VisualDSP++ 5.0 Linker and Utilities Manual 2-58 Linker Command-Line Reference

Use -flags-pp with caution. For example, if the pp legacy L comment syntax is enabled, the comment characters become unavailable for non-comment syntax.

-h[elp]

The -h or -help switch directs the assembler to output to a list of command-line switches with a syntax summary.

-i|I directory

The -idirectory or -Idirectory (include directory) switch directs the linker to append the specified directory to the search path for included files. To add multiple directories, repeat the switch or specify a list of directories terminated by semicolons (;) with the final semicolon being optional.

-ip

The -ip (individual placement) switch directs the linker to fill in fragmented memory with individual data objects that fit. When the -ip switch is specified on the linker’s command line (or via the VisualDSP++ IDDE), the default behavior of the linker—placing data blocks in consec- utive memory addresses—is overridden. The -ip switch allows individual placement of a grouping of data in processor memory to provide more efficient memory packing. Absolute placements take precedence over data/program section placements in contiguous memory locations. When remaining memory space is not sufficient for the entire section placement, the link fails. The -ip switch allows the linker to extract a block of data for individual placement and fill in fragmented memory spaces.

2-59 VisualDSP++ 5.0 Linker and Utilities Manual Linker

-jcs2l [ Used with Blackfin processors only. The -jcs2l (jump/call short to long) switch directs the linker to convert out-of-range calls and jump instructions to a code sequence that will use an indirect jump or call. Because the indirect sequence uses a register P1, the expansion will only be applied to instructions that use the CALL.X or JUMP.X opcodes. The following table shows how the Blackfin linker handles jump/call conversions.

Instruction Without -jcs2l With -jcs2l

JUMP.S Short Short

JUMP Short or long Short or long

JUMP.L Long Long

JUMP.X Short or long Short, long, or indirect

CALL CALL CALL

CALL.X CALL CALL or indirect

Refer to the instruction set reference for target architecture for more information on jump and call instructions.

-jcs2l+ [ Used with Blackfin processors only. This deprecated switch is equivalent to the -jcs2l switch.

-keep symbolName

The -keep symbolName (keep unused symbols) switch directs the linker to keep symbols from being eliminated. It directs the linker (when -e or -ev

VisualDSP++ 5.0 Linker and Utilities Manual 2-60 Linker Command-Line Reference

is enabled) to retain listed symbols in the executable even if they are unused.

-meminit

The -meminit (post-process executable file) switch directs the linker to post-process the .dxe file through the Memory Initializer utility. (For more information, see “Memory Initializer” in Chapter 7, Memory Ini- tializer.) This action causes the sections specified in the .ldf file to be run-time initialized by the C run-time library. By default, if this flag is not specified, all sections are initialized at “load” time (for example, via the VisualDSP++ IDDE or the boot loader). Refer to “SECTIONS{}” on page 3-61 for more information on section initialization. For information about the __MEMINIT predefined macro, see “__MEMINIT__” on page 3-36.

-nomemcheck

The -nomemcheck (memory checking off) switch allows you to turn off memory checking.

-o filename

The -o filename (output file) switch sets the value of the $COMMAND_LINE_OUTPUT_FILE macro which is normally used as a parameter to the LDF OUTPUT() command, which specifies the output file name. If no -o is present on command line, the $COMMAND_LINE_OUTPUT_FILE macro gets a value of “a.dxe”.

-od directory

The -od directory switch directs the linker to specify the value of the $COMMAND_LINE_OUTPUT_DIRECTORY LDF macro. This switch allows you to make a command-line change that propagates to many places without changing the LDF. Refer to “Built-In LDF Macros” on page 3-30.

2-61 VisualDSP++ 5.0 Linker and Utilities Manual Linker

-pp

The -pp (end after preprocessing) switch directs the linker to stop after the preprocessor runs without linking. The output (preprocessed LDF) is printed to a file with the same name as the .ldf file with an .is extension. This file is in the same directory as the .ldf file.

-proc processor

The -proc processor (target processor) switch directs the linker to produce code suitable for the specified processor. For example,

linker -proc ADSP-BF533 p0.doj p1.doj p2.doj -o program.dxe

See also “-si-revision version” for more information on silicon L revision of the specified processor.

-reserve-null

The -reserve-null switch directs the linker to reserve four addressable units (words) in memory at address 0x0. The switch is useful for C/C++ programs, to avoid allocation of code or data at the 0x0 (NULL pointer) address.

-s

The -s (strip all symbols) switch directs the linker to omit all symbol information from the output file.

Some debugger functionality (including “run to main”), all stdio [ functions, and the ability to stop at the end of program execution rely on the debugger’s ability to locate certain symbols in the executable file. This switch removes these symbols.

VisualDSP++ 5.0 Linker and Utilities Manual 2-62 Linker Command-Line Reference

-save-temps

The -save-temps switch directs the linker to save temporary (intermedi- ate) output files.

-si-revision version

The -si-revision version (silicon revision) switch directs the linker to build for a specific hardware revision. Any errata workarounds available for the targeted silicon revision will be enabled. The version parameter represents a silicon revision of the processor specified by the -proc switch (on page 2-62). For example,

linker -proc ADSP-BF533 -si-revision 0.1

If silicon version “none” is used, no errata workarounds are enabled. Specifying silicon version “any” enables all errata workarounds for the target processor.

If the -si-revision switch is not used, the linker builds for the latest known silicon revision for the target processor, and any errata workarounds appropriate for the latest silicon revision are enabled.

If the silicon revision is set to “any”, the __SILICON_REVISION__ macro is set to 0xffff. If the -si-revision switch is set to “none”, the linker will not set the __SILICON_REVISION__ macro.

The linker passes the -si-revision switch when invoking another VisualDSP++ tool, for example when the linker invokes the assembler. Example: The Blackfin linker invoked as

linker -proc ADSP-BF533 -si-revision 0.1 …

invokes the assembler with

2-63 VisualDSP++ 5.0 Linker and Utilities Manual Linker

easmblkfn -proc ADSP-BF533 -si-revision 0.1

-sp

The -sp (skip preprocessing) switch directs the linker to link without preprocessing the .ldf file.

-t

The -t (trace) switch directs the linker to output the names of link objects to standard output as the linker processes them.

-tx

The -tx (full trace) switch directs the linker to output the full names of link objects (full directory path) to standard output as the linker processes them.

-v[erbose]

The -v or -verbose (verbose) switch directs the linker to display version and command-line information for each phase of linking.

-version

The -version (display version) switch directs the linker to display version information for the linker.

-warnonce

The -warnonce (single symbol warning) switch directs the linker to warn only once for each undefined symbol, rather than once for each reference to that symbol.

VisualDSP++ 5.0 Linker and Utilities Manual 2-64 Linker Command-Line Reference

-xref

The -xref switch directs the linker to produce an XML cross-reference file xref.xml in the linker output directory. The XML file can be opened in a web-browser for viewing.

This linker switch is distinct from the -xref compiler driver L switch.

2-65 VisualDSP++ 5.0 Linker and Utilities Manual 3 LINKER DESCRIPTION FILE

Every DSP project requires one Linker Description File (.ldf). The .ldf file specifies precisely how to link projects. Chapter 2, “Linker”, describes the linking process and how the .ldf file ties into the linking process.

When generating a new .ldf file, use the Expert Linker to generate L an .ldf file. Refer to Chapter 4, “Expert Linker” for details.

The .ldf file allows code development for any processor system. It defines your system to the linker and specifies how the linker creates executable code for your system. This chapter describes.ldf file syntax, structure and components. Refer to Appendix C, “LDF Programming Examples for TigerSHARC Processors”, Appendix D, “LDF Programming Examples for SHARC Processors”, and Appendix E, “LDF Programming Examples for Blackfin Processors” for example .ldf files for typical systems. This chapter contains: • “LDF File Overview” on page 3-3 • “LDF File Structure” on page 3-18 • “LDF Expressions” on page 3-20 • “LDF Keywords, Commands, and Operators” on page 3-21 • “LDF Operators” on page 3-23 • “LDF Macros” on page 3-29 • “LDF Commands” on page 3-36

VisualDSP++ 5.0 Linker and Utilities Manual 3-1 The linker runs the preprocessor on the .ldf file, so you can use L preprocessor commands (such as #defines) within the file. For information about preprocessor commands, refer to a VisualDSP++ 5.0 Assembler and Preprocessor Manual. Assembler section declarations in this document correspond to the assembler’s .SECTION directive. Refer to example DSP programs shipped with VisualDSP++ for sample .ldf files supporting typical system models.

3-2 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

LDF File Overview

The .ldf file directs the linker by mapping code or data to specific memory segments. The linker maps program code (and data) within the system memory and processor(s), and assigns an address to every symbol, where:

symbol = label symbol = function_name symbol = variable_name

If you neither write an .ldf file nor import an .ldf file into your project, nor have VisualDSP++ generate an .ldf file, VisualDSP++ links the code using a default .ldf file. The chosen default .ldf file is determined by the processor specified in the VisualDSP++ Project Options dialog box. Default .ldf files are packaged with your processor tool distribution kit in a subdirectory specific to your target processor’s family. One default .ldf file is provided for each processor supported by your VisualDSP++ installation (see “Default LDFs”).

The .ldf file combines information, directing the linker to place input sections in an executable file according to the memory available in the DSP system. The linker may output warning messages and error messages. L You must resolve the error messages to enable the linker to produce valid output. See “Linker Warning and Error Messages” on page 2-10 for more information.

Blackfin-Generated LDFs

On the Blackfin platform, the VisualDSP++ New Project Wizard and the Project Options dialog allow you to generate and configure a custom Linker Description File (.ldf). (Add an .ldf file via the Add Startup Code/LDF subpage of the Project Options dialog box.) This is the quick-

VisualDSP++ 5.0 Linker and Utilities Manual 3-3 LDF File Overview

est and easiest way to customize your .ldf files. See VisualDSP++ Help for information about the Project Wizard and the Project Options dialog box.

Default LDFs

The name of each .ldf file indicates the intended processor (for example, ADSP-BF531.ldf). If the .ldf file name has no suffix, it is the “default .ldf file”. That is, when no .ldf file is explicitly specified, the default file is used to link an application when building for that processor. Therefore, ADSP-BF531.ldf is the default .ldf file for the ADSP-BF531 processor.

If no .ldf file is specified explicitly via the –T command-line switch, the compiler driver selects the default .ldf file for the target processor. For example, the first of the following commands uses the default .ldf file, and the second uses a user-specified file:

ccblkfn –proc ADSP-BF531 hello.c // uses default ADSP-BF531.ldf

ccblkfn –proc ADSP-BF531 hello.c –T ./my.ldf // uses ./my.ldf

On SHARC and TigerSHARC platforms, for each processor, there are three .ldf files with the suffixes _C, _CPP, and _ASM (for example, ADSP-21363_C.ldf).

On SHARC and TigerSHARC platforms, these .ldf files are templates for the Expert Linker. If you use the Expert Linker to create a custom .ldf file for your project, the Expert Linker queries for the kind (assembly, C, or C++) of .ldf file you want to create and then copies one of the above templates. The suffixes indicate the kind of .ldf files they support.

3-4 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

The CPP template is a superset of the C template, and the C template is a superset of the ASM template. The differences are as follows:

• The CPP template links against C++ run-time libraries, C++ exception libraries, and the run-time headers built to initialize C++ constructors. It maps data sections that contain information controlling how thrown exceptions are caught.

• The C template is currently identical to the CPP template, since a C project may link against local or system libraries that have been implemented in C++. There may be differences in a future release.

• The ASM template does not include a run-time header, and does not permit command-line arguments to applications. The ASM template is not suitable for use with profile-guided optimization. Since the ASM template has no run-time header, it does not mandate a “start” symbol resolved to the Reset address. It does not map the C++ exception sections into memory.

Each .ldf file handles a variety of demands, allowing applications to be built in multiple configurations, merely by supplying a few command-line options. This flexibility is achieved by extensive use of preprocessor macros within the .ldf file. Macros serve as flags to indicate one choice or another, and as variables within the .ldf file to hold the name of a chosen file or other link-time parameter. This reliance on preprocessor operation can make the .ldf file seem an imposing sight. In simple terms, different LDF configurations are selected by defining preprocessor macros on the linker command line. This can be specified from the Link page of the VisualDSP++ IDDE’s Project Options dialog box or directly from the command line.

At the top of the default Blackfin .ldf files, you will find documentation on the macros you can use to configure the default .ldf files.

VisualDSP++ 5.0 Linker and Utilities Manual 3-5 LDF File Overview

You can use an .ldf file written from scratch. However, modifying an existing .ldf file (or a default .ldf file) is often the easier alternative when there are no large changes in your system’s hardware or software. See Listing 3-1 on page 3-7, Listing 3-2 on page 3-10, and Listing 3-3 on page 3-11 for examples of basic .ldf files for supported processors. See “Common Notes on Basic LDF Examples” on page 3-13 for basic information on LDF structure. See “LDF Programming Examples for TigerSHARC Processors” on page E-1, “LDF Programming Examples for SHARC Processors” on page D-1, and “LDF Programming Examples for Blackfin Processors” on page C-1 for code examples for TigerSHARC, SHARC, and Blackfin processors, respectively.

3-6 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

Example 1 – Basic LDF for Blackfin Processors

Listing 3-1 is an example of a basic .ldf file for ADSP-BF535 processors (formatted for readability). Note the MEMORY{} and SECTIONS{} commands and refer to “Common Notes on Basic LDF Examples” on page 3-13. Other LDF examples are provided in “LDF Programming Examples for Blackfin Processors”.

Listing 3-1. Example LDF for ADSP-BF535 Processor

ARCHITECTURE(ADSP-BF535) SEARCH_DIR($ADI_DSP/Blackfin/lib) $OBJECTS = CRT, $COMMAND_LINE_OBJECTS ENDCRT;

MEMORY /* Define/label system memory */ { /* List of global Memory Segments */ MEM_L2 { TYPE(RAM) START(0xF0000000) END(0xF002FFFF) WIDTH(8) } MEM_HEAP { TYPE(RAM) START(0xF0030000) END(0xF0037FFF) WIDTH(8) } MEM_STACK { TYPE(RAM) START(0xF0038000) END(0xF003DFFF) WIDTH(8) } MEM_SYSSTACK { TYPE(RAM) START(0xF003E000) END(0xF003FDFF) WIDTH(8) } MEM_ARGV { TYPE(RAM) START(0xF003FE00) END(0xF003FFFF) WIDTH(8) } }

PROCESSOR P0 { /* the only processor in the system */ OUTPUT ( $COMMAND_LINE_OUTPUT_FILE )

SECTIONS { /* List of sections for processor P0 */

VisualDSP++ 5.0 Linker and Utilities Manual 3-7 LDF File Overview

L2 { INPUT_SECTION_ALIGN(2) /* Align all code sections on 2 byte boundary */ INPUT_SECTIONS( $OBJECTS(program) $LIBRARIES(program)) INPUT_SECTION_ALIGN(1) INPUT_SECTIONS( $OBJECTS(data1) $LIBRARIES(data1)) INPUT_SECTION_ALIGN(1) INPUT_SECTIONS( $OBJECTS(constdata) $LIBRARIES(constdata)) INPUT_SECTION_ALIGN(1) INPUT_SECTIONS( $OBJECTS(ctor) $LIBRARIES(ctor) ) } >MEM_L2

stack { ldf_stack_space = .; ldf_stack_end = ldf_stack_space + MEMORY_SIZEOF(MEM_STACK) - 4; } >MEM_STACK

heap { /* Allocate a heap for the application */ ldf_heap_space = .; ldf_heap_end = ldf_heap_space + MEMORY_SIZEOF(MEM_HEAP) - 1; ldf_heap_length = ldf_heap_end - ldf_heap_space; } >MEM_HEAP

argv { /* Allocate argv space for the application */ ldf_argv_space = .; ldf_argv_end =

3-8 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

ldf_argv_space + MEMORY_SIZEOF(MEM_ARGV) - 1; ldf_argv_length = ldf_argv_end - ldf_argv_space; } >MEM_ARGV

} /* end SECTIONS */

} /* end PROCESSOR p0 */

Memory Usage in Blackfin Processors

The default .ldf files define memory areas for all defined spaces on the processor.1 Not all of these memory areas are used within the .ldf files. Instead, the .ldf files provide three basic memory configurations: • The default configuration specifies that only internal memory is available and caching is disabled. Thus, no code or data is mapped to SDRAM unless explicitly placed there, and all of the available L1 space is used for code or data.

• Defining the USE_CACHE macro selects the alternative configuration, where code and data caches are enabled and external SDRAM is used. Code and data are mapped into L1 where possible, but the Cache/SRAM areas are left empty; any spill-over goes into the SDRAM.

• Defining the USE_SDRAM macro has the same effect as defining the USE_CACHE macro, except that code and data are mapped to the L1 Cache/SRAM areas.

If USE_CACHE is used, caches may safely be turned on, because doing so will not corrupt code or data. Selecting this option does not actually enable the caches — that must be done separately (for example, through the

1 With the exception of the core MMRs, which the linker considers “out of bounds”.

VisualDSP++ 5.0 Linker and Utilities Manual 3-9 LDF File Overview

___cplb_ctrl configuration variable). Instead, this option ensures that the memory layout allows caches to be enabled later. A common user error occurs when cache is enabled despite not having specified USE_CACHE. This leads to code or data corruption as cache activity overwrites the contents of SRAM. Therefore, the LDFs use the following “guard symbols”:

___l1_code_cache

___l1_data_cache_a

___l1_data_cache_b

These symbols are defined by the .ldf files and are given values (that is, resolved to addresses 0 or 1), depending on whether USE_CACHE is defined. The run-time library examines these symbols when cache configuration is requested, and refuses to enable a cache if the corresponding guard symbol is zero, indicating that valid information already occupies this space. For more information, refer to VisualDSP++ 5.0 C/C++ Compiler and Library Manual, section “Caching and Memory Protection”.

Example 2 – Basic LDF for TigerSHARC Processors

Listing 3-2 is an example of a basic .ldf file for the ADSP-TS101 processor (formatted for readability). Note the MEMORY{} and SECTIONS{} commands and refer to “Common Notes on Basic LDF Examples” on page 3-13. Other LDF examples are provided in “LDF Programming Examples for TigerSHARC Processors”.

Listing 3-2. Example LDF for ADSP-TS201 Processor

ARCHITECTURE(ADSP-TS101) SEARCH_DIR($ADI_DSP/TS/lib) $OBJECTS = main.doj, $COMMAND_LINE_OBJECTS;

3-10 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

MEMORY { /* Define and label system memory */ /* List of global memory segments */ M0Code {TYPE(RAM) START(0x000000) END(0x00FFFF) WIDTH(32)} M1Data {TYPE(RAM) START(0x080000) END(0x08FFFF) WIDTH(32)} M2Data {TYPE(RAM) START(0x100000) END(0x10FFFF) WIDTH(32)} } PROCESSOR P0 { /* the only processor in the system */ OUTPUT ( $COMMAND_LINE_OUTPUT_FILE ) SECTIONS{ code { INPUT_SECTIONS ( $OBJECTS(program))} > M0Code data1 { INPUT_SECTIONS ( $OBJECTS(data1))} > M1Data data2 { INPUT_SECTIONS ( $OBJECTS(data2))} > M2Data } /* End of SECTIONS command for processor P0 */ } /* End of PROCESSOR command. /*

Example 3 – Basic LDF for SHARC Processors

Listing 3-3 is an example of a basic .ldf file for the ADSP-21161 processor (formatted for readability). Note the MEMORY{} and SECTIONS{} commands and refer to “Common Notes on Basic LDF Examples” on page 3-13. Other examples for assembly and C source files are in “LDF Programming Examples for SHARC Processors”.

Listing 3-3. Example LDF File for ADSP-21161 Processor

// Link for the ADSP-21161 ARCHITECTURE(ADSP-21161) SEARCH_DIR ( $ADI_DSP/211xx/lib ) MAP (SINGLE-PROCESSOR.XML) // Generate a MAP file

// $ADI_DSP is a predefined linker macro that expands to // the VisualDSP++ installation directory. Search for objects // in directory 21k/lib relative to the installation directory

VisualDSP++ 5.0 Linker and Utilities Manual 3-11 LDF File Overview

// lib161.dlb is an ADSP-2116x-specific library and must precede // precede libc.dlb, C library to link 2116x-specific routines

$LIBS = lib161.dlb, libc.dlb;

// single.doj is a user-generated file. // The linker will be invoked as follows: // linker -T single-processor.ldf single.doj. // $COMMAND_LINE_OBJECTS is a predefined linker macro. // The linker expands this macro into the name(s) of the // the object(s) (.doj files) and libraries (.dlb files) // that appear on the command line. In this example, // $COMMAND_LINE_OBJECTS = single.doj

// 161_hdr.doj is the standard initialization file for 2116x $OBJS = $COMMAND_LINE_OBJECTS, 161_hdr.doj;

// A linker project to generate a .dxe file PROCESSOR P0 { OUTPUT ( ./SINGLE.dxe ) // The name of the output file

MEMORY // Processor-specific memory command { INCLUDE("21161_memory.h"}

SECTIONS // Specify the output sections { INCLUDE("21161_sections.h" ) } // end P0 sections } // end P0 processor

3-12 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

Common Notes on Basic LDF Examples

In the following description, the MEMORY{} and SECTIONS{} commands connect the program to the target processor. For syntax information on LDF commands, see “LDF Commands” on page 3-36.

These notes describe features of a typical .ldf file (as presented in Listing 3-1, Listing 3-2, and Listing 3-3).

• ARCHITECTURE(ADSP-xxxxx) specifies the target architecture (processor). For example, ARCHITECTURE(ADSP-BF533). The architecture dictates possible memory widths and address ranges, the register set, and other structural information used by the debugger, linker, and loader. The target architecture must be installed in VisualDSP++.

• SEARCH_DIR() specifies directory paths searched for libraries and object files (on page 3-60). For example, the argument $ADI_DSP/Blackfin/lib specifies one search directory for Blackfin libraries and object files. The linker supports a sequence of search directories presented as an argument list (directory1, directory2, ...). The linker follows this sequence and stops at the first match.

• $LIBRARIES is a list of the library and object files searched to resolve references, in the required order. Some of the options specify the selection of one library over another.

• $OBJECTS is an example of a user-definable macro, which expands to a comma-delimited list of file names. Macros improve readability by replacing long strings of text. Conceptually similar to preprocessor macro support (#defines) also available in the .ldf file, string macros are independent. In this example, $OBJECTS expands to a comma-delimited list of the input files to be linked.

VisualDSP++ 5.0 Linker and Utilities Manual 3-13 LDF File Overview

Note: In this example and in the default .ldf files that accompany VisualDSP++, $OBJECTS in the SECTIONS() command specifies the object files to be searched for specific input sections.

As another example, $ADI_DSP expands to the VisualDSP++ home directory.

• $COMMAND_LINE_OBJECTS (on page 3-30) is an LDF command-line macro, which expands in the .ldf file into the list of input files that appears on the command line. Note: The order in which the linker processes object files (which affects the order in which addresses in memory segments are assigned to input sections and symbols) is determined by the order the files are listed in INPUT_SECTIONS() commands. As noted above, this order is typically the order listed in $OBJECTS ($COMMAND_LINE_OBJECTS). VisualDSP++ generates a linker command line that lists objects in alphabetical order. This order carries through to the $OBJECTS macro. You may customize the .ldf file to link objects in any desired order. Instead of using default macros such as $OBJECTS, each INPUT_SECTION command can have one or more explicit object names.

3-14 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

The following examples are functionally identical:

dxe_program { INPUT_SECTIONS ( main.doj(program)

fft.doj(program) ) } > mem_program

$DOJS = main.doj, fft.doj;

dxe_program {

INPUT_SECTIONS ($DOJS(program))

} >mem_program;

• The MEMORY{} command (on page 3-44) defines the target system’s physical memory and connects the program to the target system. Its arguments partition the memory into memory segments. Each memory segment is assigned a distinct name, memory type, a start and end address (or segment length), and a memory width. These names occupy different namespaces from input section names and output section names. Thus, a memory segment and an output section may have the same name.

• Each PROCESSOR{} command (on page 3-54) generates a single executable file.

• The OUTPUT() command (on page 3-54) produces an executable (.dxe) file and specifies its file name.

In the basic example, the argument to the OUTPUT() command is the $COMMAND_LINE_OUTPUT_FILE macro (on page 3-30). The linker names the executable file according to the text following the -o

VisualDSP++ 5.0 Linker and Utilities Manual 3-15 LDF File Overview

switch (which corresponds to the name specified in the Project Options dialog box when the linker is invoked via the VisualDSP++ IDDE).

linker ... -o outputfilename

• SECTIONS{} (on page 3-61) specifies the placement of code and data in physical memory. The linker maps input sections (in object files) to output sections (in executable files), and maps the output sections to memory segments specified by the MEMORY{} command.

• The INPUT_SECTIONS() statement specifies the object file that the linker uses as an input to resolve the mapping to the appropriate memory segment declared in the .ldf file. • For example, in TigerSHARC processors, the following INPUT_SECTIONS() statement directs the linker to place the program input section in the code output section and to map it to the M0Code memory segment.

code { INPUT_SECTIONS ( $OBJECTS(program))} > M0Code

• For SHARC processors, the following INPUT_SECTIONS() statement directs the linker to place the isr_tbl input section in the dxe_isr output section and to map it to the mem_isr memory segment.

dxe_isr{ INPUT_SECTIONS ( $OBJECTS (isr_tbl) ) } > mem_isr

• For Blackfin processors, the following two input sections (program and data1) are mapped into one memory segment (L2), as shown below.

dxe_L2

1 INPUT_SECTIONS_ALIGN (2)

3-16 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

2 INPUT_SECTIONS($OBJECTS(program) $LIBRARIES(program))

3 INPUT_SECTIONS_ALIGN (1)

4 INPUT_SECTIONS($OBJECTS(data1) $LIBRARIES(data1))

}>MEM_L2

The second line directs the linker to place the object code assembled from the source file’s “program” input section (via the “.section program” directive in the assembly source file), place the output object into the “DXE_L2” output section, and map the output section to the “MEM_L2” memory segment. The fourth line does the same for the input section “data1” and output section “DXE_L2”, mapping them to the memory segment “MEM_L2”. The two pieces of code follow each other in the program memory segment.

The INPUT_SECTIONS() commands are processed in the same order as object files appear in the $OBJECTS macro. You may intersperse INPUT_SECTIONS() statements within an output section with other directives, including location counter information.

VisualDSP++ 5.0 Linker and Utilities Manual 3-17 LDF File Structure

LDF File Structure

One way to produce a simple and maintainable .ldf file is to parallel the structure of your DSP system. Using your system as a model, follow these guidelines.

• Split the file into a set of PROCESSOR{} commands, one for each DSP in your system.

• Place a MEMORY{} command in the scope that matches your system and define memory unique to a processor within the scope of the corresponding PROCESSOR{} command.

• If applicable, place a SHARED_MEMORY{} command in the .ldf file’s global scope. This command specifies system resources available as shared resources in a multiprocessor environment. Declare common (shared) memory definitions in the global scope before the PROCESSOR{} commands. See “Command Scoping” for more information. Comments in the LDF

C-style comments begin with /* and may cross “newline” boundaries until a */ terminator is encountered.

A C++ style comment begins with // and ends at the end of the line.

For more information on .ldf file structure, see: • “Link Target Description” on page 2-11 • “Placing Code on the Target” on page 2-36 Also see “LDF Programming Examples for TigerSHARC Processors” on page E-1, “LDF Programming Examples for SHARC Processors” on

3-18 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

page D-1, and “LDF Programming Examples for Blackfin Processors” on page C-1 for code and .ldf file structure examples for TigerSHARC, SHARC, and Blackfin processors, respectively.

Command Scoping

The two LDF scopes are global and command (see Figure 3-1).

MEMORY{} MPMEMORY{} SHARED_MEMORY Scope of SHARED_MEMORY{} { OUTPUT() Global SECTIONS{} LDF } Scope PROCESSOR P0 { OUTPUT() Scope of PROCESSOR P0{} MEMORY{} SECTIONS{} RESOLVE{} } Figure 3-1. LDF Command Scoping Example

A global scope occurs outside commands. Commands and expressions that appear in the global scope are always available and are visible in all subsequent scopes. LDF macros are available globally, regardless of the scope in which the macro is defined (see “LDF Macros” on page 3-29). A command scope applies to all commands that appear between the braces ({ }) of another command, such as a PROCESSOR{} or PLIT{} command. Commands and expressions that appear in the command scopes are lim- ited to those scopes.

Figure 3-1 illustrates some scoping issues. For example, the MEMORY{} command that appears in the LDF’s global scope is available in all com-

VisualDSP++ 5.0 Linker and Utilities Manual 3-19 LDF Expressions

mand scopes, but the MEMORY{} command that appears in command scopes is restricted to those scopes.

LDF Expressions

LDF commands may contain arithmetic expressions that follow the same syntax rules as C/C++ language expressions. The linker:

• Evaluates all expressions as type unsigned long and treats constants as type unsigned long • Supports all C/C++ language arithmetic operators • Allows definitions and references to symbolic constants in the LDF • Allows reference to global variables in the program being linked • Recognizes labels that conform to these constraints: • Must start with a letter, an underscore, or point • May contain any letters, underscores, digits, or points • Are delimited by white space • Do not conflict with any keywords • Are unique Table 3-1 lists valid items used in expressions.

Table 3-1. Valid Items in Expressions

Convention Description . Current location counter (a period character in an address expression). See “Location Counter (.)” on page 3-29.

0xnumber Hexadecimal number (a 0x prefix)

3-20 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

Table 3-1. Valid Items in Expressions (Cont’d)

Convention Description

number Decimal number (a number without a prefix)

numberk A decimal number multiplied by 1024 or numberK

B#number A binary number or b#number

LDF Keywords, Commands, and Operators

Descriptions of LDF keywords, operators, macros, and commands are provided in the following sections. • “LDF Keywords” on page 3-22 • “Miscellaneous LDF Keywords” on page 3-23 • “LDF Operators” on page 3-23 • “LDF Macros” on page 3-29 • “Built-in Preprocessor Macros” on page 3-33 • “LDF Commands” on page 3-36 Keywords are case sensitive; the linker recognizes a keyword only L when the entire word is UPPERCASE.

VisualDSP++ 5.0 Linker and Utilities Manual 3-21 LDF Keywords, Commands, and Operators

LDF Keywords

Table 3-2 lists all general LDF keywords (used in Blackfin, SHARC, and TigerSHARC processor families).

Table 3-2. LDF Keywords Summary

ABSOLUTE ADDR ALGORITHM

ALIGN ALL_FIT ARCHITECTURE

BEST_FIT BOOT COMMON_MEMORY

DEFINED DYNAMIC ELIMINATE

ELIMINATE_SECTIONS ENTRY END

FALSE FILL FIRST_FIT

INCLUDE INPUT_SECTION_ALIGN INPUT_SECTIONS

INPUT_SECTIONS_PIN INPUT_SECTIONS_PIN_ KEEP EXCLUSIVE

KEEP_SECTIONS LENGTH LINK_AGAINST

MAP MEMORY MEMORY_SIZEOF

MPMEMORY NUMBER_OF_OVERLAYS OUTPUT

OVERLAY_GROUP OVERLAY_ID OVERLAY_INPUT

OVERLAY_OUTPUT PACKING PLIT

PLIT_SYMBOL_ADDRESS PLIT_SYMBOL_OVERLAYID PROCESSOR

RAM RESERVE RESOLVE

RESERVE_EXPAND ROM

SEARCH_DIR SECTIONS SHARED_MEMORY

SHT_NOBITS SIZE SIZEOF

START TYPE DATA64

VERBOSE WIDTH XREF

PM DM SW

3-22 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

Miscellaneous LDF Keywords

The following linker keywords are not operators, macros, or commands.

Table 3-3. Miscellaneous LDF Keywords

Keyword Description

FALSE A constant with a value of 0

TRUE A constant with a value of 1

XREF A cross-reference option setting. See “-xref” on page 2-65.

For more information about other LDF keywords, see “LDF Operators” on page 3-23, “LDF Macros” on page 3-29, and “LDF Commands” on page 3-36.

LDF Operators

LDF operators in expressions support memory address operations. Expressions that contain these operators terminate with a semicolon, except when the operator serves as a variable for an address. The linker responds to several LDF operators including the location counter. Each LDF operator is described in the following sections.

ABSOLUTE() Operator Syntax:

ABSOLUTE(expression)

VisualDSP++ 5.0 Linker and Utilities Manual 3-23 LDF Keywords, Commands, and Operators

The linker returns the value expression. Use this operator to assign an absolute address to a symbol. The expression can be: • A symbolic expression in parentheses; for example,

ldf_start_expr = ABSOLUTE(start + 8);

This example assigns ldf_start_expr the value corresponding to the address of the symbol start, plus 8, as in:

ldf_start_expr = start + 8;

• An integer constant in one of these forms: hexadecimal, decimal, or decimal optionally followed by “K” (kilo [x1024]) or “M” (Mega [x1024x1024]) • A period, indicating the current location (see “Location Counter (.)” on page 3-29) The following statement, which defines the bottom of stack space in the LDF

ldf_stack_space = .;

can also be written as:

ldf_stack_space = ABSOLUTE(.);

• A symbol name

ADDR() Operator Syntax:

ADDR(section_name) This operator returns the start address of the named output section defined in the .ldf file. Use this operator to assign a section’s absolute address to a symbol.

3-24 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

Blackfin Code Example:

If an .ldf file defines output sections as,

dxe_L2_code { INPUT_SECTIONS( $OBJECTS(program) $LIBRARIES(program)) }> mem_L2

dxe_L2_data { INPUT_SECTIONS( $OBJECTS(data1) $LIBRARIES(data1)) }> mem_L2

the .ldf file may contain the command:

ldf_start_L2 = ADDR(dxe_L2_code)

The linker generates the constant ldf_start_L2 and assigns it the start address of the dxe_L2 output section. SHARC Code Example:

If an .ldf file defines output sections as,

dxe_pmco { INPUT_SECTIONS( $OBJECTS(seg_pmco) $LIBRARIES(seg_pmco)) }> mem_pmco

dxe_dmda { INPUT_SECTIONS( $OBJECTS(seg_dmda) $LIBRARIES(seg_dmda)) }> mem_seg_dmda

the .ldf file may contain the command:

ldf_start_dmda = ADDR(mem_seg_dmda)

VisualDSP++ 5.0 Linker and Utilities Manual 3-25 LDF Keywords, Commands, and Operators

The linker generates the constant ldf_start_dmda and assigns it the start address of the mem_seg_dmda output section.

DEFINED() Operator Syntax:

DEFINED(symbol)

The linker returns 1 when the symbol appears in the global symbol table, and returns 0 when the symbol is not defined. Use this operator to assign default values to symbols. Example:

If an assembly object linked by the .ldf file defines the global symbol test, the following statement sets the test_present constant to 1. Otherwise, the constant has the value 0.

test_present = DEFINED(test);

MEMORY_END() Operator Syntax:

MEMORY_END(segment_name)

This operator returns the end address (the address of the last word) of the named memory segment. Example:

This example reserves six words at the end of a mem_stack memory segment using the MEMORY_END operator. RESERVE(reserved_space = MEMORY_END(mem_stack) – 6 + 1, reserved_space_length = 6)

3-26 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

MEMORY_SIZEOF() Operator Syntax:

MEMORY_SIZEOF(segment_name)

This operator returns the size (in words) of the named memory segment. Use this operator when a segment’s size is required to move the current location counter to an appropriate memory location. Example:

This example (from a default .ldf file) sets a linker-generated constant based on the location counter plus the MEMORY_SIZEOF operator.

sec_stack { ldf_stack_limit = .; ldf_stack_base = . + MEMORY_SIZEOF(mem_stack) - 1; } > mem_stack

The sec_stack section is defined to consume the entire mem_stack memory segment.

MEMORY_START() Operator Syntax:

MEMORY_START(segment_name)

This operator returns the start address (the address of the first word) of the named memory segment. Example:

This example reserves four words at the start of a mem_stack memory segment using the MEMORY_START operator: RESERVE(reserved_space = MEMORY_START(mem_stack), reserved_space_length = 4)

VisualDSP++ 5.0 Linker and Utilities Manual 3-27 LDF Keywords, Commands, and Operators

The sec_stack section is defined to consume the entire mem_stack memory segment.

SIZEOF() Operator Syntax:

SIZEOF(section_name)

This operator returns the size (in bytes) of the named output section. Use this operator when a section’s size is required to move the current location counter to an appropriate memory location. SHARC Code Example:

The following code fragment defines the _sizeofdata1 constant to the size of the seg_dmda section.

seg_dmda { INPUT_SECTIONS( $OBJECTS(seg_dmda) $LIBRARIES(seg_dmda)) _sizeofdata1 = SIZEOF(seg_dmda); } > seg_dmda

Blackfin Code Example:

The following code fragment defines the _sizeofdata1 constant to the size of the data1 section.

data1 { INPUT_SECTIONS( $OBJECTS(data1) $LIBRARIES(data1)) _sizeofdata1 = SIZEOF(data1); } > MEM_DATA1

3-28 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

Location Counter (.)

The linker treats a “.” (period surrounded by spaces) as the symbol for the current location counter. The location counter is a pointer to the memory location at the end of the previous linker command. Because the period refers to a location in an output section, this operator may appear only within an output section in a SECTIONS{} command. Observe these rules: • Use a period anywhere a symbol is allowed in an expression. • Assign a value to the period operator to move the location counter and to leave voids or gaps in memory. • Do not allow the location counter to be decremented.

LDF Macros

LDF macros (or linker macros) are built-in macros. They have predefined system-specific procedures or values. Other macros, called user macros, are user-definable.

LDF macros are identified by a leading dollar sign ($) character. Each LDF macro is a name for a text string. You may assign LDF macros with textual or procedural values, or simply declare them to exist. The linker: • Substitutes the string value for the name. Normally, the string value is longer than the name, so the macro expands to its textual length. • Performs actions conditional on the existence of (or value of) the macro • Assigns a value to the macro, possibly as the result of a procedure, and uses that value in further processing

VisualDSP++ 5.0 Linker and Utilities Manual 3-29 LDF Keywords, Commands, and Operators

LDF macros funnel input from the linker command line into predefined macros and provide support for user-defined macro substitutions. Linker macros are available globally in the .ldf file, regardless of where they are defined. For more information, see “Command Scoping” on page 3-19 and “LDF Macros and Command-Line Interaction” on page 3-32. LDF macros are independent of preprocessor macro support, L which is also available in the .ldf file. The preprocessor places preprocessor macros (or other preprocessor commands) into source files. Preprocessor macros (see “Built-in Preprocessor Macros” on page 3-33) repeat instruction sequences in your source code or define symbolic constants. These macros facilitate text replacement, file inclusion, and conditional assembly and compilation. For example, the assembler’s preprocessor uses the #define command to define macros and symbolic constants. For more information, refer to the VisualDSP++ 5.0 Compiler and Library Manual for appropriate target processor and the VisualDSP++ 5.0 Assembler and Preprocessor Manual.

Built-In LDF Macros The linker provides the following built-in LDF macros.

• $COMMAND_LINE_OBJECTS

This macro expands into the list of object (.doj) and library (.dlb) files that are input on the linker’s command line. Use this macro within the INPUT_SECTIONS() syntax of the

3-30 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

linker’s SECTIONS{} command. This macro provides a comprehensive list of object file input that the linker searches for input sections.

• $COMMAND_LINE_LINK_AGAINST

This macro expands into the list of executable (.dxe or .sm) files that one input on the linker’s command line. This macro provides a comprehensive list of executable file input that the linker searches to resolve external symbols.

• $COMMAND_LINE_OUTPUT_FILE This macro expands into the output executable file name, which is set with the linker’s -o switch. This file name corresponds to the set via the VisualDSP++ Project Options dialog box. Use this macro only once in your LDF for file name substitution within an OUTPUT() command.

• $COMMAND_LINE_OUTPUT_DIRECTORY This macro expands into the path of the output directory, which is set with the linker’s -od switch (or -o switch when -od is not specified).

VisualDSP++ 5.0 Linker and Utilities Manual 3-31 LDF Keywords, Commands, and Operators

For example, the following statement permits a configuration change (release vs. debug) without modifying the .ldf file.

OVERLAY_OUTPUT($COMMAND_LINE_OUTPUT_DIRECTORY/OVL1.ovl)

• $ADI_DSP This macro expands into the path of the VisualDSP++ installation directory. Use this macro to control how the linker searches for files.

User-Declared Macros The linker supports user-declared macros for file lists. The following syntax declares $macroname as a comma-delimited list of files. $macroname = file1, file2, file3, ... ;

After $macroname has been declared, the linker substitutes the file list when $macroname appears in the .ldf file. Terminate a $macroname declaration with a semicolon. The linker processes the files in the listed order.

LDF Macros and Command-Line Interaction The linker receives commands through a command-line interface, regardless of whether the linker runs automatically from the VisualDSP++ IDDE or explicitly from a command window. Many linker operations, such as input and output, are controlled through command-line entries. Use LDF macros to apply command-line inputs within the .ldf file.

3-32 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

Base your decision on whether to use command-line inputs in the .ldf file or to control the linker with LDF code on the following considerations.

•An .ldf file that uses command-line inputs produces a more generic .ldf file that can be used in multiple projects. Because the command line can specify only one output, an .ldf file that relies on command-line input is best suited for single-processor systems.

•An .ldf file that does not use command-line inputs produces a more specific .ldf file that can control complex linker features.

Built-in Preprocessor Macros

The linker’s preprocessor defines a number of macros to provide information about the linker. These macros can be tested, using the #ifdef and related directives, to support your program’s needs. This section provides information about the following built-in preprocessor macros. • __VISUALDSPVERSION__ • __VERSIONNUM__ • __VERSION__ • __SILICON_REVISION__ • __MEMINIT__

__VISUALDSPVERSION__

The __VISUALDSPVERSION__ predefined macro provides VisualDSP++ product version information. The macro allows a pre-processing check to be placed within the .ldf file. It can be used to differentiate between VisualDSP++ releases and updates. This macro applies to all Analog Devices processors.

VisualDSP++ 5.0 Linker and Utilities Manual 3-33 LDF Keywords, Commands, and Operators

Syntax:

__VISUALDSPVERSION__=0xMMmmUUxx

Table 3-4 explains the parameters of this macro.

Table 3-4. __VISUALDSPVERSION Macro Parameters

Parameter Description MM VersionMajor. The major release number; for example, 4 in release 4.5. mm VersionMinor. The minor release number; for example, 5 in release 4.5. UU VersionPatch. The number of the release update, such as version 4.5, update 6. xx Reserved for future use (always 00 initially)

The 0xMMmmUUxx information is obtained from the \Sys- tem\VisualDSP.ini file. The xx is initially set at 00.

If an unexpected problem occurs in trying to locate VisualDSP.ini or in extracting information from the VisualDSP.ini file, the __VISUALDSPVERSION__ macro will not be encoded to the VisualDSP++ product version. The __VISUALDSPVERSION__ 0xffffffff string is displayed as part of an error message when the version information is unable to be encoded. Code Example (Legacy):

#if !defined(__VISUALDSPVERSION__) #warning Building with VisualDSP++ 4.5 Update 5 or prior. No __VISUALDSPVERSION__ available. #endif

Code Example (VisualDSP++ 4.5 Update 6 or Later):

#if __VISUALDSPVERSION__ >= 0x04050600 #warning Building with VisualDSP++ 4.5 Update 6 or later #endif

3-34 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

Code Example (Error Check):

#if __VISUALDSPVERSION__ == 0xffffffff #error Unexpected build problems, unknown VisualDSP++ Version #endif

__VERSIONNUM__

The __VERSIONNUM__ predefined macro provides VisualDSP++ linker version information in hex form. The macro allows a pre-processing check to be placed within the .ldf file. It can be used to differentiate between VisualDSP++ linker versions. This macro applies to all Analog Devices processors.

In other words, this macro defines __VERSIONNUM__ as a numeric variant of __VERSION__ constructed from the version number of the linker. Eight bits are used for each component in the version number and the most significant byte of the value represents the most significant version component.

For example, a linker with version 3.6.0.0 defines __VERSIONNUM__ as 0x03060000 and 3.6.2.10 would define __VERSIONNUM__ to be 0x0306020A.

__VERSION__

The __VERSION__ predefined macro provides VisualDSP++ linker version information in string form, giving the version number of the linker. The macro allows a pre-processing check to be placed within the .ldf file. It can be used to differentiate between VisualDSP++ linker versions. This macro applies to all Analog Devices processors. For example, for linker version 3.9.1.1, the value of the macro would be 3.9.1.1.

VisualDSP++ 5.0 Linker and Utilities Manual 3-35 LDF Keywords, Commands, and Operators

__SILICON_REVISION__

The __SILICON_REVISION__ predefined macro value is defined by the -si-revision version switch.

For example, if the silicon revision switch (-si-revision) is set to “any”, the __SILICON_REVISION__ macro is set to 0xffff. If the -si-revision switch is set to “none”, the linker does not set the __SILICON_REVISION__ macro.

__MEMINIT__

The __MEMINIT__ predefined macro is defined if the -meminit switch is used on the command line.

LDF Commands

Commands in the .ldf file (called LDF commands) define the target system and specify the order in which the linker processes output for that system. LDF commands operate within a scope, influencing the operation of other commands that appear within the range of that scope. For more information, see “Command Scoping” on page 3-19. The linker supports the use of wildcards in section name specifica- L tions in the .ldf file. The * and ? are provided on input section names. The linker supports these LDF commands (not all commands are used with specific processors): • “ALIGN()” on page 3-37 • “ARCHITECTURE()” on page 3-38 • “COMMON_MEMORY{}” on page 3-38 • “ELIMINATE()” on page 3-39

3-36 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

• “ELIMINATE_SECTIONS()” on page 3-40 • “INCLUDE()” on page 3-40 • “INPUT_SECTION_ALIGN()” on page 3-40 • “KEEP()” on page 3-42 • “KEEP_SECTIONS()” on page 3-42 • “LINK_AGAINST()” on page 3-42 • “MEMORY{}” on page 3-44 • “MPMEMORY{}” on page 3-47 • “OVERLAY_GROUP{}” on page 3-48 • “PACKING()” on page 3-48 • “PLIT{}” on page 3-53 • “PROCESSOR{}” on page 3-54 • “RESERVE()” on page 3-56 • “RESERVE_EXPAND()” on page 3-58 • “RESOLVE()” on page 3-59 • “SEARCH_DIR()” on page 3-60 • “SECTIONS{}” on page 3-61 • “SHARED_MEMORY{}” on page 3-72

ALIGN()

The ALIGN(number) command aligns the address of the current location counter to the next address that is a multiple of number, where number is a

VisualDSP++ 5.0 Linker and Utilities Manual 3-37 LDF Keywords, Commands, and Operators

power of 2. The number is a word boundary (address) that depends on the word size of the memory segment in which the ALIGN() takes action.

ARCHITECTURE()

The ARCHITECTURE() command specifies the target system’s processor. An ..ldf file may contain one ARCHITECTURE() command only. The ARCHITECTURE() command must appear with global LDF scope, applying to the entire .ldf file. The command’s syntax is:

ARCHITECTURE(processor)

The ARCHITECTURE() command is case sensitive. For example, a valid entry is ADSP-BF533. Thus, ADSP-BF533 is valid, but adsp-BF533 is not.

If the ARCHITECTURE() command does not specify the target processor, you must identify the target processor via the linker command line (linker -proc processor ...). Otherwise, the linker cannot link the program.

If processor-specific MEMORY{} commands in the .ldf file conflict with the processor type, the linker issues an error message and halts. Test whether your VisualDSP++ installation accommodates L a particular processor by typing the following linker command. linker -proc processor

If the architecture is not installed, the linker prints a message to that effect.

COMMON_MEMORY{}

The COMMON_MEMORY{} command is used to map objects into memory that is shared by more than one processor. The mapping is done in the context

3-38 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

of the processors that will use the shared memory; these processors are identified as a "master" of the common memory. For detailed command description, refer to “COMMON_MEMORY{}” on page 5-53.

ELIMINATE()

The ELIMINATE() command enables object elimination, which removes symbols from the executable file if they are not called. Adding the VERBOSE keyword, ELIMINATE(VERBOSE), reports on objects as they are eliminated. This command performs the same function as the -e command-line switch (see on page 2-57). When using either the linker’s data elimination feature (via the Expert Linker or command-line switches) or the ELIMINATE() command in an .ldf file, it is essential that certain objects are continue to use the KEEP() command, so that the C/C++ run-time libraries function properly. The safest way to do this is to copy the KEEP() command from the default .ldf file into your own .LDF file. For the C and C++ run-time libraries to work properly, retain the L following symbols with “KEEP()” (on page 3-42): ___ctor_NULL_marker and ___lib_end_of_heap_descriptions. In order to allow efficient elimination, the structure of the assembly source has to be such that the linker can unambiguously identify the boundaries of each “source object” in the input section (a “source object” is a function or a data item). Specifically, an input section must be fully covered by non-overlapping source objects with explicit boundaries. The boundary of a function item is specified by the function label and its corresponding “.end” label. If an input section layout does not conform to the rule described above, no elimination is performed in the section. See the VisualDSP++ 5.0 Assembler and Preprocessor Manual for more details on using “.end” labels.

VisualDSP++ 5.0 Linker and Utilities Manual 3-39 LDF Keywords, Commands, and Operators

ELIMINATE_SECTIONS()

The ELIMINATE_SECTIONS(sectionList) command instructs the linker to remove unreferenced code and data from listed sections only.

The sectionList is a comma-delimited list of input sections. Both this LDF command and the linker’s -es command-line switch (on page 2-58) may be used to specify sections where unreferenced code and data should be eliminated.

ENTRY()

The ENTRY(symbol) command specifies the entry address. The entry address is usually filled from a global symbol “start” (no underscore), if present. Refer to “Entry Address” on page 2-34 for more information.

Both this LDF command and the linker’s -entry command-line switch (on page 2-58) may be used to specify the entry address.

INCLUDE()

The INCLUDE() command specifies additional .ldf files that the linker processes before processing the remainder of the current .ldf file. Specify any number of additional .ldf files. Supply one file name per INCLUDE() command.

Only one of these additional .ldf files is obligated to specify a target architecture. Normally, the top-level .ldf files includes the other .ldf files.

INPUT_SECTION_ALIGN()

The INPUT_SECTION_ALIGN(number) command aligns each input section (data or instruction) in an output section to an address satisfying number. The number argument, which must be a power of 2, is a word boundary

3-40 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

(address). Valid values for number depend on the word size of the memory segment receiving the output section being aligned.

The linker fills empty spaces created by INPUT_SECTION_ALIGN() commands with zeros (by default), or with the value specified with the preceding FILL command valid for the current scope. See FILL under “SECTIONS{}” on page 3-61.

The INPUT_SECTION_ALIGN() command is valid only within the scope of an output section. For more information, see “Command Scoping” on page 3-19. For more information on output sections, see the syntax description for “SECTIONS{}” on page 3-61. Example:

In the following Blackfin example, input sections from a.doj, b.doj, and c.doj are aligned on even addresses. Input sections from d.doj and e.doj are not quad-word aligned because INPUT_SECTION_ALIGN(1) indicates subsequent sections are not subject to input section alignment.

SECTIONS { program { INPUT_SECTION_ALIGN(2)

INPUT_SECTIONS ( a.doj(program)) INPUT_SECTIONS ( b.doj(program)) INPUT_SECTIONS ( c.doj(program))

// end of alignment directive for input sections INPUT_SECTION_ALIGN(1)

// The following sections will not be aligned. INPUT_SECTIONS ( d.doj(data1)) INPUT_SECTIONS ( e.doj(data1))

VisualDSP++ 5.0 Linker and Utilities Manual 3-41 LDF Keywords, Commands, and Operators

} >MEM_PROGRAM }

KEEP()

The linker uses the KEEP(keepList) command when section elimination is enabled, retaining the listed objects in the executable file even when they are not called. The keepList is a comma-delimited list of objects to be retained. When utilizing the linker’s data elimination capabilities, it is essential that certain objects continue to use the KEEP() command, so that the C/C++ run-time libraries function properly. The safest way to do this is to copy the KEEP() command from the default .ldf file into your own .ldf file. For the C and C++ run-time libraries to work properly, retain the L following symbols with KEEP: ___ctor_NULL_marker and ___lib_end_of_heap_descriptions A symbol specified in keeplist must be a global symbol.

KEEP_SECTIONS()

The linker uses the KEEP_SECTIONS() command to specify a section name in which elimination should not take place. This command can appear anywhere the ELIMINATE_SECTION command appears. You may either use the KEEP_SECTIONS() command or the -ek switch (on page 2-57).

LINK_AGAINST()

The LINK_AGAINST() command checks specific executables to resolve variables and labels that have not been resolved locally.

To link programs for multiprocessor systems, a LINK_AGAINST() L command must be present in the .ldf file.

3-42 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

This command is an optional part of the PROCESSOR{} and SHARE_MEMORY{} commands. The syntax of the LINK_AGAINST() command (as part of a PROCESSOR{} command) is:

PROCESSOR Pn { ... LINK_AGAINST (executable_file_names) ... }

where:

• Pn is the processor name; for example, P0 or P1.

• executable_file_names is a list of one or more executable (.dxe) or shared memory (.sm) files. Separate multiple file names with commas. However, Expert Linker allows the use of white spaces to separate multiple file names. The linker searches the executable files in the order specified in the LINK_AGAINST() command. When a symbol’s definition is found, the linker stops searching. Override the search order for a specific variable or label by using the RESOLVE() command (see “RESOLVE()” on page 3-59), which directs the linker to use the specified resolver, thus ignoring LINK_AGAINST() for a specific symbol. The LINK_AGAINST() command for other symbols still applies.

MAP()

The MAP(filename) command outputs a map (.xml) file with the specified name. You must supply the file name. Place this command anywhere in the .ldf file.

The MAP(filename) command corresponds to (and may be overridden by) the linker’s -Map command-line switch (on page 2-55). In VisualDSP++, if project options (Link page of the Project Options dialog

VisualDSP++ 5.0 Linker and Utilities Manual 3-43 LDF Keywords, Commands, and Operators

box) specify the generation of a symbol map, the linker runs with -Map .xml asserted and the .ldf file’s MAP() command generates a warning.

MEMORY{}

The MEMORY{} command specifies the memory map for the target system. After declaring memory segment names with this command, use the memory segment names to place program sections via the SECTIONS{} command.

The .ldf file must contain a MEMORY{} command for global memory on the target system and may contain a MEMORY{} command that applies to each processor’s scope. There is no limit to the number of memory segments you can declare within each MEMORY{} command. For more information, see “Command Scoping” on page 3-19.

In each scope scenario, follow the MEMORY{} command with a SECTIONS{} command. Use the memory segment names to place program sections. Only memory segment declarations may appear within the MEMORY{} command. There is no limit to section name lengths. If you do not specify the target processor’s memory map with the MEMORY{} command, the linker cannot link your program. If the combined sections directed to a memory segment require more space than exists in the segment, the linker issues an error message and halts the link.

3-44 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

The syntax for the MEMORY{} command appears in Figure 3-2, followed by a description of each part of a segment declaration.

MEMORY{ segment_commands }

segment_name { TYPE(RAM | ROM ) START(address_expression ) LENGTH( length_expression) | END(address_expression) WIDTH(width_expression ) } Figure 3-2. MEMORY{} Command Syntax Tree

Segment Declarations

A segment declaration declares a memory segment on the target processor. Although an .ldf file may contain only one MEMORY{} command that applies to all scopes, there is no limit to the number of memory segments declared within a MEMORY{} command.

Each segment declaration must contain a segment_name, TYPE(), START(), LENGTH() or END(), and a WIDTH(). The parts of a segment declaration are described below. segment_name

The segment_name identifies the memory region. The segment_name must start with a letter, underscore, or point, may include any letters, underscores, digits, and points, and must not conflict with LDF keywords.

START(address_number)

The START() command specifies the memory segment’s start address. The address_number must be an absolute address.

VisualDSP++ 5.0 Linker and Utilities Manual 3-45 LDF Keywords, Commands, and Operators

TYPE()

The TYPE() command identifies the architecture-specific type of memory within the memory segment. Not all target processors support all types of memory. The linker L stores this information in the executable file for use by other development tools.

For Blackfin and TigerSHARC processors, use TYPE() to specify the functional or hardware locus (RAM or ROM). The RAM declarator specifies segments that need to be booted. ROM segments are not booted; they are executed/loaded directly from off-chip PROM space.

For SHARC (ADSP-21xxx) processors, use TYPE()to specify two parameters: memory usage (PM for program memory or DM for data memory), and functional or hardware locus (RAM or ROM, as described above). On ADSP-21261/2/6/7 and ADSP-21362/3/4/5/6 processors, it is not possible to access external memory directly, but through DMA. To vali- date placement of code accessible through DMA in external memory, use the DMAONLY segment qualifier to mark a memory segment in the .ldf file as external memory. For example,

seg_dmda { TYPE(DM DMAONLY) START(0x00200000) END(0x3FFFFFFF) WIDTH(32) } <…> seg_dmda{INPUT_SECTIONS( $OBJECTS(seg_extm) )} > seg_dmda

The linker identifies the section as dmaonly. At link time, the linker verifies that the section must reside in external memory identified with the DMAONLY qualifier. More importantly, the linker checks that only

3-46 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

sections marked dmaonly are placed in external memory. The linker issues an error if there is any inconsistency between memory the section is mapped to and that section’s qualifier:

[Error el2017] Invalid/missing memory qualifier for memory 'section name.

LENGTH(length_number)/END(address_number)

The LENGTH/END() command identifies the length of the memory segment (in words) or specifies the segment’s end address. When you state the length, length_number is the number of addressable words within the region. When you state the end address, address_number is an absolute address.

WIDTH(width_number)

The WIDTH() command specifies the physical width (number of bits) of the on-chip or off-chip memory interface. The width_number parameter must be a whole number. The parameters are: • For Blackfin processors, width must be 8 (bits) • For TigerSHARC processors, width must be 32 (bits) • For SHARC processors, width may be 8, 16, 32, 48, or 64 (bits)

MPMEMORY{}

The MPMEMORY{} command specifies the offset of each processor’s physical memory in a multiprocessor target system. After you declare the processor names and memory segment offsets with the MPMEMORY{} command, the linker uses the offsets during multiprocessor linking. Refer to “MPMEMORY{}” on page 5-45 for a detailed description of the MPMEMORY{} command.

VisualDSP++ 5.0 Linker and Utilities Manual 3-47 LDF Keywords, Commands, and Operators

OVERLAY_GROUP{}

The OVERLAY_GROUP{} command is deprecated. This command provides support for defining a set of overlays that share a block of run-time memory. For detailed command description, refer to “OVERLAY_GROUP{}” on page 5-29. Refer to “Memory Management Using Overlays” on page 5-4 for a detailed description of overlay functionality.

PACKING()

In VisualDSP++ 5.0, the PACKING() command is used with L ADSP-21xxx (SHARC) processors only (as described in “Packing in SHARC Processors” on page 3-50). Processors exchange data with their environment (on-chip or off-chip) through several buses. The configuration, placement, and amounts of memory are determined by the application. Specify memory of width(s) and data transfer byte order(s) that suit your needs. The linker places data in memory according to the constraints imposed by your system’s architecture. The LDF PACKING() command specifies the order the linker uses to place bytes in memory. This ordering places data in memory in the sequence the processor uses as it transfers data.

The PACKING() command allows the linker to structure its executable output to be consistent with your installation’s memory organization. This command can be applied (scoped) on a segment-by-segment basis within the .ldf file, with adequate granularity to handle heterogeneous memory configurations. Any memory segment requiring more than one packing command may be divided into homogeneous segments. Syntax

The syntax of the PACKING() command is:

3-48 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

PACKING (number_of_bytes byte_order_list)

where:

• number_of_bytes is an integer specifying the number of bytes to pack (reorder) before repeating the pattern

• byte_order_list is the output byte ordering – what the linker writes into memory. Each list entry consists of “B” followed by the byte’s number (in a group) at the storage medium (memory).

The list follows these rules: • Parameters are whitespace-delimited

• The total number of non-null bytes is number_of_bytes

• If null bytes are included, they are labeled B0

For example, in SHARC processors, the first byte is B1 (not B0). The second byte is B2, and so on. PACKING (12 B1 B2 B3 B4 B0 B11 B12 B5 B6 B0 B7 B8 B9 B10 B0)

Non-default use of the PACKING() command reorders bytes in executable files (.dxe, .sm, or .ovl), so they arrive at the target in the correct number, alignment, and sequence. To accomplish this task, the command specifies the size of the reordered group, the byte order within the group, and whether and where “null” bytes must be inserted to preserve alignment on the target. The term “null” refers to usage – the target ignores a null byte; the linker sets these bytes to zeros. The order used to place bytes in memory correlates to the order the processor may use while unpacking the data when the processor transfers data from external memory into its internal memory. The processor’s unpacking order can relate to the transfer method.

VisualDSP++ 5.0 Linker and Utilities Manual 3-49 LDF Keywords, Commands, and Operators

VisualDSP++ comes with the packing.h file in the .../include L folder. This file provides macros that define packing commands for use in an LDF. The macros support various types of packing for direct memory access functionality (used in overlays) and for direct external execution. To use these macros, place them in an .ldf file’s SECTIONS{} command when a PACKING() command is needed.

Packing in SHARC Processors

On SHARC processors, PACKING() applies to the processor’s external port. Each external port buffer contains data packing logic that allows the packing of 8-, 16-, or 32-bit external bus words into 32- or 48-bit internal words. This logic is fully reversible.

The following information describes how the PACKING() command may apply in an .ldf file for your ADSP-21xxx processor. In some direct memory access (DMA) modes, SHARC processors unpack three 32-bit words to build two 48-bit instruction words when the processor receives data from 32-bit memory. For example, the unpacked order and storage order (Table 3-5) could apply to a DMA mode.

Table 3-5. DMA Packing Order

Transfer Order Unpacked Order (from storage in a 32-bit external memory) Two 48-bit internal words (after the third transfer) B1 and B2 (word 1, bits 47-32) B3 and B4 (word 1, bits 31-16) B1, B2, B3, B4, B5, B6 B11 and B12 (word 2, bits 15-0) (word 1, bits 47-0) B5 and B6 (word 1, bits 15-0) B7, B8, B9, B10, B11, B12 B7 and B8 (word 2, bits 47-32) (word 2, bits 47-0) B9 and B10 (word 2, bits 31-16)

3-50 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

The order of unpacked bytes does not match the transfer (stored) order. Because the processor uses two bytes per short word, the above transfer translates into the format in Table 3-6.

Table 3-6. Storage Order vs. Unpacked Order

Storage Order Unpacked Order (in 32-bit external memory) (two 48-bit internal words) B1, B2, B3, B4, B11, B12 B1, B2, B3, B4, B5, B6 B5, B6, B7, B8, B9, B10 B7, B8, B9, B10, B11, B12

You specify to the linker how to accommodate processor-specific byte packing (for example, non-sequential byte order) with the PACKING() syntax within the OVERLAY_INPUT{} command. The above example’s byte ordering translates into the following PACKING() command syntax, which supports 48-bit to 32-bit packing over the processor’s external port. PACKING (12 B1 B2 B3 B4 B0 B11 B12 B5 B6 B0 B7 B8 B9 B10 B0)

The above PACKING() syntax places instructions in an overlay stored in a 32-bit external memory, but is unpacked and executed from 48-bit internal memory.

Refer to fft_ovly.fft, which uses a macro that defines the packing. This file is included with the overlay3 example that ships with VisualDSP++.

Overlay Packing Formats in SHARC Processors

Use the PACKING() command when: • Data and instructions for overlays are executed from external memory (by definition those overlays “live” in external memory) • The width or byte order of stored data differs from its run-time organization The linker word-aligns the packing instruction as needed.

VisualDSP++ 5.0 Linker and Utilities Manual 3-51 LDF Keywords, Commands, and Operators

Table 3-7 indicates packing format combinations for SHARC DMA overlays available under each of the two operations. Table 3-8 indicates packing format combinations for ADSP-21161N overlays available for storage in 8-bit-wide memory; 8-bit packing is available on ADSP-2106x and ADSP-21160 processors during EPROM booting only.

Table 3-7. Packing Formats for SHARC DMA Overlays

Execution Storage Memory Memory type type Packing Instruction

32-bit PM 16-bit DM PACKING(6 B0 B0 B1 B2 B5 B0 B0 B3 B4 B6)

32-bit DM 16-bit DM PACKING(4 B0 B0 B1 B2 B0 B0 B3 B4 B5)

48-bit PM 16-bit DM PACKING(6 B0 B0 B1 B2 B0 B0 B0 B3 B4 B0 B0 B0 B5 B6 B0)

48-bit DM 32-bit DM PACKING(12 B1 B2 B3 B4 B0 BB5 B6 B11 B12 B0 B7 B8 B9 B10 B0)

Table 3-8. Additional Packing Formats for DMA Overlays

Execution Storage Memory type Memory type Packing Instruction

48-bit PM 8-bit DM PACKING(6 B0 B0 B0 B1 B0 B0 B0 B2 B0 B0 B0 B3 B0 B0 B0 B0 B4 B0 B0 B0 B0 B5 B0 B0 B0 B0 B6 B0 B0 B0 B0 B0 B0 B0 B0 B0 B0 B0)

32-bit DM 8-bit DM PACKING(4 B0 B0 B0 B1 B0 B0 B0 B0 B2 B0 B0 B0 B0 B3 B0 B0 B0 B0 B4 B0)

16-bit DM 8-bit DM PACKING(2 B0 B0 B0 B1 B0 B0 B0 B0 B2 B0)

External Execution Packing in SHARC Processors The only two processors that require packed memory for external execution are the ADSP-21161N and the ADSP-21065L chips. The ADSP-21161N processor supports 48-, 32-, 16-, and 8-bit-wide external

3-52 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

memory. The ADSP-21065L processor supports 32-bit external memory only. Previous to VisualDSP++ 3.5, it was required to use “packing” L commands in the .ldf file to cause the code to be placed properly. In VisualDSP++ 3.5 and latter releases, the VisualDSP++ tools are enhanced to perform packing automatically. In order for the VisualDSP++ tools to execute packing directly from external memory on ADSP-21065L and ADSP-21161N processors, the tools “pack” the code into the external memory providing the following conditions are met: 1. Ensure the “type” of the external memory is PM (Program Memory) 2. Ensure the data width matches the “real/actual” memory width: ADSP-21065L processors – 32 bits; ADSP-21161N processors – 48, 32, 16, and 8 bits

3. If the .ldf file has the PACKING() command for the particular section, remove the command. When defining memory segments (required for external memory), the “type” of a memory section is recommended to be: • PM – code or 40-bit data (data requires PX register to access) • DM – all other sections Width should be the “actual/physical” width of the external memory.

PLIT{}

The PLIT{} (procedure linkage table) command in an .ldf file inserts assembly instructions that handle calls to functions in overlays. The PLIT{} commands provide a template from which the linker generates assembly code when a symbol resolves to a function in overlay memory.

VisualDSP++ 5.0 Linker and Utilities Manual 3-53 LDF Keywords, Commands, and Operators

Refer to “PLIT{}” on page 5-34 for a detailed description of the PLIT{} command. Refer to “Memory Management Using Overlays” on page 5-4 for a detailed description of overlay and PLIT functionality.

PROCESSOR{}

The PROCESSOR{} command declares a processor and its related link information. A PROCESSOR{} command contains the MEMORY{}, SEC- TIONS{}, RESOLVE{}, and other linker commands that apply only to that specific processor.

The linker produces one executable file from each PROCESSOR{} command. If you do not specify the type of link with a PROCESSOR{} command, the linker cannot link your program.

The syntax for the PROCESSOR{} command appears in Figure 3-3.

PROCESSOR processor_name { OUTPUT(file_name.DXE) [MEMORY{segment_commands }] [PLIT{plit_commands } ] SECTIONS{section_commands} RESOLVE(symbol, resolver) }

Figure 3-3. PROCESSOR{} Command Syntax Tree

3-54 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

The PROCESSOR{} command syntax is defined as:

• processor_name Assigns a name to the processor. Processor names follow the same rules as linker labels. For more information, see “LDF Expressions” on page 3-20.

• OUTPUT(file_name.dxe)

Specifies the output file name for the executable (.dxe) file. An OUTPUT() command in a scope must appear before the SECTIONS{} command in that same scope.

• MEMORY{segment_commands} Defines memory segments that apply only to this specific processor. Use command scoping to define these memory segments outside the PROCESSOR{} command. For more information, see “Command Scoping” on page 3-19 and “MEMORY{}” on page 3-44.

• PLIT{plit_commands} Defines procedure linkage table (PLIT) commands that apply only to this specific processor. For more information, see “PLIT{}” on page 3-53.

• SECTIONS{section_commands}

Defines sections for placement within the executable (.dxe) file. For more information, see “SECTIONS{}” on page 3-61.

• RESOLVE{symbol, resolver}

Ignores any LINK_AGAINST() command. For details, see the “RESOLVE()” command.

VisualDSP++ 5.0 Linker and Utilities Manual 3-55 LDF Keywords, Commands, and Operators

Multiprocessor/Multicore Applications The PROCESSOR{} command may be used in linking projects on multiprocessor/multicore Blackfin architectures such as the ADSP-BF561 processor. For example, the command syntax for two-processor system is as follows:

PROCESSOR p0 { ... } PROCESSOR p1 { }

See also “LINK_AGAINST()” on page 3-42, “MPMEMORY{}” on page 5-45, “COMMON_MEMORY{}” on page 5-53, and “SHARED_MEMORY{}” on page 5-47.

RESERVE()

The RESERVE (start_symbol, length_symbol, min_size [,align]) command allocates address space and defines symbols start_symbol and length_symbol. The command allocates the largest free memory block available, larger than or equal to min_size. Given an optional parameter align, RESERVE allocates aligned address space. Input:

• The min_size parameter defines a required minimum size of memory to allocate.

• The align parameter is optional and defines alignment of allocated address space.

3-56 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

Output:

• The start_symbol is assigned the starting address of the allocated address space.

• The length_symbol is assigned the size of the allocated address space.

A user may restrict the command by defining the start and length symbols together or individually. For example,

RESERVE (start_symbol = address, length_symbol, min_size) RESERVE (start_symbol = address, length_symbol = size) RESERVE (start_symbol, length_symbol = size [,align])

The RESERVE() command is valid only within the scope of an output section. For more information on output sections, see “Command Scoping” on page 3-19 and “SECTIONS{}” on page 3-61. Also see “RESERVE_EXPAND()” on page 3-58 for more information on how to claim any unused memory after input sections have been mapped.

Linker Error Resolutions

Linker error li1224: When a user defines length_symbol, the min_size parameter is redundant and not included in the command. When a user defines start_symbol, the align parameter is redundant and not included in the command.

Linker errors li1221, li1222, and li1223: When a user defines start_symbol = address, the align parameter is redundant and should not be included in the command.

When a user defines align parameter, the length_symbol or min_size parameter should be divisible by align; the align parameter must be a power of 2.

Given the start_symbol is not restricted (not defined), RESERVE allocates address space, starting from a segment end address.

VisualDSP++ 5.0 Linker and Utilities Manual 3-57 LDF Keywords, Commands, and Operators

Example Consider an example where given memory segment [0 - 8]. Range [0 - 2] is used by an input section. To allocate address space of minimum size 4 and aligned by 2, the RESERVE command has minimum length requirement of 4 and alignment 2.

M0 {START(0), END(8), WIDTH(1)}

out{… RESERVE(start, length, 4, 2) } >M0

1. Allocate 4 words {5, 6, 7, 8},

start = 5

length = 4

2. To satisfy alignment by 2, allocate address space {4, 5, 6, 7, 8}

start = 4

length = 5

3. Consider length exactly 4 (not minimum 4). Allocated address space is {4, 5, 6, 7}. Address [8] is freed.

start = 4

length = 4

RESERVE_EXPAND()

The RESERVE_EXPAND(start_symbol, length_symbol, min_size) command may follow a RESERVE command and is used to define same symbols as RESERVE. Ordinarily, RESERVE_EXPAND is specified last in an output section to claim any unused memory after input sections have been mapped. RESERVE_EXPAND attempts to allocate memory adjacent to the range allocated by RESERVE. Accordingly, start_symbol and

3-58 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

length_symbol are redefined to include expanded address range. Refer to “RESERVE()” on page 3-56 for more information.

RESOLVE()

Use the RESOLVE(symbol_name, resolver) command to ignore a LINK_AGAINST() command for a specific symbol. This command overrides the search order for a specific variable or label. Refer to “LINK_AGAINST()” on page 3-42 for more information.

The RESOLVE(symbol_name, resolver) command uses the resolver to specify an address of a particular symbol (variable or label). The resolver is an absolute address or a file (.dxe or .sm) that contains the symbol’s definition. For example,

RESOLVE(start, 0xFFA00000)

If the symbol is not located in the designated file, an error is issued.

For the RESOLVE(symbol_name, resolver) command: • When the symbol is not defined in the current processor scope, the supplies a file name, overriding any LINK_AGAINST(). • When the symbol is defined in the current processor scope, the supplies to the linker the symbol location address. Resolve a C variable by prefixing the variable with an underscore in L the RESOLVE() command (for example, _symbol_name).

Potential Problem with Symbol Definition

Assume the symbol used in the RESOLVE() command is defined in the link project. The linker will use that definition from the link project rather one from the symbol_name, resolver) (also known as “resolve-against”) link project specified in the RESOLVE() command. For example,

VisualDSP++ 5.0 Linker and Utilities Manual 3-59 LDF Keywords, Commands, and Operators

RESOLVE(_main, p1.dxe) linker -T a.ldf -Map a.map -o .\Debug\a.dxe

The linker then issues the following message:

[Warning li2143] "a.ldf":12 Symbol '_main' used in resolve-against command is defined in processor 'p0'.

If you want to use a local definition, remove the RESOLVE() command. Otherwise, remove the definition of the symbol from the link project.

SEARCH_DIR()

The SEARCH_DIR() command specifies one or more directories that the linker searches for input files. Specify multiple directories within a SEARCH_DIR command by delimiting each path with a semicolon (;). On Windows, enclose long directory names with embedded spaces within straight quotes. The search order follows the order of the listed directories. This command appends search directories to the directory selected with the linker’s -L command-line switch (on page 2-54). Place this command at the beginning of the .ldf file to ensure that the linker applies the command to all file searches. Example:

ARCHITECTURE (ADSP-Blackfin) MAP (SINGLE-PROCESSOR.XML) // Generate a MAP file

SEARCH_DIR( $ADI_DSP/Blackfin/lib; ABC/XYZ ) // $ADI_DSP is a predefined linker macro that expands // to the VisualDSP++ install directory. Search for objects // in directory Blackfin/lib relative to the install directory // and to the ABC/XYZ directory.

3-60 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

SECTIONS{}

The SECTIONS{} command uses memory segments (defined by MEMORY{} commands) to specify the placement of output sections into memory. Figure 3-4 shows syntax for the SECTIONS{} command.

SECTIONS{ section_statements}

expression section_name [type_qualifier|init_qualifier]{section_commands }[> memory_segment]

FORCE_CONTIGUITY|NO_FORCE_CONTIGUITY

INPUT_SECTIONS(file_source {archive_member }(input_labels) )

LDF macro expression list_of_files

OVERLAY_OUTPUT(file_name.OVL) INPUT_SECTIONS(input_section_commands) FILL(hex number) ALGORITHM(ALL_FIT) SIZE(expression) RESOLVE_LOCALLY(TRUE|FALSE)

PLIT{ plit_commands}

OVERLAY_INPUT(overlay_commands)>overlay_live_memory_segment

Figure 3-4. SECTIONS{} Command Syntax Tree

An .ldf file may contain one SECTIONS{} command within each of the PROCESSOR{} commands. The SECTIONS{} command must be preceded by a MEMORY{} command, which defines the memory segments in which the linker places the output sections. Though an .ldf file may contain only one SECTIONS{} command within each processor command scope, multiple output sections may be declared within each SECTIONS{} command.

The SECTIONS{} command’s syntax includes several arguments.

VisualDSP++ 5.0 Linker and Utilities Manual 3-61 LDF Keywords, Commands, and Operators

expressions or section_declarations

Use expressions to manipulate symbols or to position the current location counter. Refer to “LDF Expressions” on page 3-20.

Use a section_declaration to declare an output section. Each section_declaration has a section_name, optional section_type or init_qualifier, section_commands, and a memory_segment.

Parts of a SECTION declaration are:

• section_name Starts with a letter, underscore, or period and may include any letters, underscores, digits, and points. A section_name must not conflict with any LDF keywords.

The special section name .PLIT indicates the procedure linkage table (PLIT) section that the linker generates when resolving symbols in overlay memory. Place this section in non-overlay memory to manage references to items in overlay memory.

• type_qualifier Specifies the address space into which the section should be mapped and the logical organization of the data. Note that this qualifier applies only to SHARC ADSP-2146x/2147x/2148x processors. The qualifiers are:

• PM – Program memory, contains 6 bytes per word.

• DM – Data memory, contains 4 bytes per word.

• DATA64 – Contains 8 bytes per word.

• SW – Contains 2 bytes per word.

3-62 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

The output section memory type supersedes the memory type that L the section is mapped into. If the output section memory type differs from the segment type, an additional ELF section is created in the output. This ELF section contains the output section and defines its contents.

The use of an output section qualifier also instructs the linker to ignore input sections whose memory type is different than specified by the qualifier. All ignored input sections from a particular mapping command are listed in the linker log file.

• init_qualifier Specifies run-time initialization type (optional). The qualifiers are:

• NO_INIT – Contains un-initialized data. There is no data stored in the .dxe file for this section (equivalent to SHT_NOBITS legacy qualifier).

• ZERO_INIT – Contains only “zero-initialized” data. If invoked with the -meminit switch (on page 2-61), the “zeroing” of the section is done at runtime by the C run-time library. If -meminit is not specified, the “zeroing” is done at “load” time.

• RUNTIME_INIT – If the linker is invoked with the -meminit switch, this section fills at runtime. If -meminit is not specified, the section fills at “load” time.

• section_commands May consist of any combination of commands and/or expressions, such as: “INPUT_SECTIONS()” on page 3-64 “expression” on page 3-69 “FILL(hex number)” on page 3-69

VisualDSP++ 5.0 Linker and Utilities Manual 3-63 LDF Keywords, Commands, and Operators

“PLIT{plit_commands}” on page 3-69 “OVERLAY_INPUT{overlay_commands}” on page 3-70 “FORCE_CONTIGUITY/NOFORCE_CONTIGUITY” on page 3-72

• memory_segment Declares that the output section is placed in the specified memory segment.

INPUT_SECTIONS()

The INPUT_SECTIONS() portion of a section_command identifies the parts of the program to place in the executable file. When placing an input section, you must specify the file_source. Optionally, you may also specify a filter expr. When file_source is a library, specify the input section’s archive_member and input_labels. The command syntax is:

INPUT_SECTIONS(library.dlb [ member.doj (input_label) ]) L Note that spaces are significant in this syntax. In the INPUT_SECTIONS() of the LDF command:

• file_source may be a list of files or an LDF macro that expands into a file list, such as $COMMAND_LINE_OBJECTS. Delimit the list of object files or library files with commas.

• archive_member names the source-object file within a library. The archive_member parameter and the left/right brackets ([]) are required when the file_source of the input_label is a library.

• input_labels are derived from run-time .SECTION names in assembly programs (for example, program). Delimit the list of names with spaces. The * and ? wildcard characters can be used to place

3-64 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

multiple section names from an object in a library. For more information about wildcard characters, see “Wildcard Characters” on page 2-35. Example:

To place the section “program” of the object “foo.doj” in the library “myLib.dlb”:

INPUT_SECTIONS(myLib.dlb [ foo.doj (program) ])

To use a wildcard character that places all sections with a prefix of “data” of the object “foo.doj” in the library “myLib.dlb”:

INPUT_SECTIONS(myLib.dlb [ foo.doj (data*) ])

Using an Optional Filter Expression The filter operation is done with curly braces, and can be used to define sub-lists and sub-libraries. It can be used for linking with attributes.

INPUT_SECTIONS( $FILES { expr } (program) )

The optional filter expr is a Boolean expression that may contain: • Attribute operators:

• name Returns true if the object has one or more attributes called name, regardless of value; otherwise, returns false.

• name("string") Returns true if the attribute name has a value that matches string. The comparison is case-sensitive string. This operator may be used on multi-valued attributes. Note that string must be quoted.

VisualDSP++ 5.0 Linker and Utilities Manual 3-65 LDF Keywords, Commands, and Operators

• name cmp-op "string" Returns true if the attribute name has a single value that matches string, according to cmp-op. Otherwise, returns false. Cmp-op can be “==” or “!=”, for equality and inequal- ity, via case-sensitive string comparison. Note that string must be quoted. This operator may only be used on single-valued attributes. If the attribute does not have exactly one value, the linker generates an error.

• name cmp-op number Returns true if the attribute name has a single value that numerically matches integer number (which can be nega- tive). Otherwise, returns false. Cmp-op can be “==”, “!=”, “<”, “<=”, “>” or “>=”. This operator may only be used on single-valued attributes. If the attribute does not have exactly one value, the linker generates an error.

• Logical operators: “&&”, “||”, and “!”, having the usual C mean- ings and precedence.

• Parentheses, for grouping: “(" and “)” Example:

$OBJS_1_and_2 = $OBJS {attr1 && attr2 }; $OBJS_3_and_2 = $OBJS { attr3("value3") && attr2 == "value2" };

Outsec { INPUT_SECTIONS($OBJS_1_and_2(program)) INPUT_SECTIONS($OBJS_3_and_2(program)) INPUT_SECTIONS($OBJS_2 { attr2 } (program)) } >mem

3-66 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

INPUT_SECTIONS_PIN/_PIN_EXCLUSIVE Commands

The INPUT_SECTIONS_PIN and INPUT_SECTIONS_PIN_EXCLUSIVE commands are used to allow mapping of an input section in one of several output sections, as in “one input section to many output section” linker feature. For example,

os_mem1 { INPUT_SECTIONS($OBJECTS(program)) } > mem1

os_mem2 { INPUT_SECTIONS($OBJECTS(program)) } > mem2

In the above example, if some of the input sections included in $OBJECTS(program) do not fit in os_mem1, the linker will try to map them into os_mem2.

An input section listed in an INPUT_SECTIONS_PIN() command will not be mapped by any INPUT_SECTIONS commands that appear later in the .ldf file, and an input section listed in INPUT_SECTIONS_PIN_EXCLUSIVE command(s) will not be mapped by any other INPUT_SECTIONS command.

Each time an input sections is mentioned in an INPUT_SECTIONS command, the linker is instructed to “give another chance” to the input section by trying to map it in different output section (given the section has not been already mapped), thus achieving the effect of “one-to-many” mapping.

The INPUT_SECTIONS_PIN() and INPUT_SECTIONS_PIN_EXCLUSIVE() commands limit the effect of “one-to-many” mapping – once the input section is mentioned inside INPUT_SECTIONS_PIN(), the linker will not map it in any of the following output sections; an input section mentioned inside INPUT_SECTIONS_PIN_EXCLUSIVE() command can not be mapped in any other output section.

VisualDSP++ 5.0 Linker and Utilities Manual 3-67 LDF Keywords, Commands, and Operators

The commands help to avoid breaking existing LDF macros. To achieve the same affect without using INPUT_SECTIONS_PIN and INPUT_SECTIONS_PIN_EXCLUSIVE commands, the definition of the output sections would have be:

os_mem1 { INPUT_SECTIONS(b.doj(program)) INPUT_SECTIONS(c.doj(program) d.doj(program)) } > mem1

os_mem2 { INPUT_SECTIONS(c.doj(program) d.doj(program)) INPUT_SECTIONS(a.doj(program)) } > mem2

Without the use of general LDF macros and INPUT_SECTIONS_PIN L commands, the .ldf file will have to change every time the list of objects changes. If the same section is mentioned in more than one of INPUT_SECTIONS_PIN() commands, linker will honor the first command only. In conjunction with attribute expressions, the commands can be used to control the order of input section placement without explicitly mention- ing the object files.

os_internal { INPUT_SECTIONS_PIN($OBJECTS{high_priority}(program)) INPUT_SECTIONS($OBJECTS(program)) } > mem_internal

os_external { INPUT_SECTIONS($OBJECTS(program)) INPUT_SECTIONS_EXCLUSIVE($OBJECTS{low_priority}(program)) } > mem_external

3-68 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

In the above example,

•“program” input sections from input files marked with “high_priority” attribute can be mapped to “mem_internal” only

•“program” input sections from input files marked with “low_priority” attribute can be mapped to “mem_external” only

• All other “program” input section can be mapped to “mem_internal” or “mem_external”

expression

In a section_command, an expression manipulates symbols or positions the current location counter. See “LDF Expressions” on page 3-20 for details.

FILL(hex number)

In a section_command, the FILL() command fills gaps (created by aligning or advancing the current location counter) with hexadecimal numbers. L The FILL() command is used only within a section declaration. By default, the linker fills gaps with zeros. Specify only one FILL() command per output section. For example,

FILL (0x0) or

FILL (0xFFFF)

PLIT{plit_commands}

In a section_command, a PLIT{} command declares a locally-scoped procedure linkage table (PLIT). It contains its own labels and expressions. For more information, see “PLIT{}” on page 5-34.

VisualDSP++ 5.0 Linker and Utilities Manual 3-69 LDF Keywords, Commands, and Operators

OVERLAY_INPUT{overlay_commands}

In a section_command, OVERLAY_INPUT{} identifies the parts of the program to place in an overlay executable (.ovl) file. For more information on overlays, see “Memory Management Using Overlays” on page 5-4. For overlay code examples, see the examples that came bundled with the development software.

The overlay_commands item consists of at least one of the following commands: INPUT_SECTIONS(), OVERLAY_ID(), NUMBER_OF_OVERLAYS(), OVERLAY_OUTPUT(), ALGORITHM(), or SIZE().

The overlay_memory_segment item (optional) determines whether the overlay section is placed in an overlay memory segment. Some overlay sections, such as those loaded from a host, do not need to be included in the overlay memory image of the executable file, but are required for other tools that read the executable file. Omitting an overlay memory segment assignment from a section retains the section in the executable file, but marks the section for exclusion from the overlay memory image of the executable file.

3-70 VisualDSP++ 5.0 Linker and Utilities Manual Linker Description File

The overlay_commands portion of an OVERLAY_INPUT{} command follows these rules.

• DEFAULT_OVERLAY When the DEFAULT_OVERLAY command is used, the linker initially places the overlay in the run-time space (that is, without running the overlay manager).

• OVERLAY_OUTPUT() Outputs an overlay (.OVL) file for the overlay with the specified name. The OVERLAY_OUTPUT() in an OVERLAY_INPUT{} command must appear before any INPUT_SECTIONS() for that overlay.

• INPUT_SECTIONS() Has the same syntax within an OVERLAY_INPUT{} command as when it appears within an output_section_command, except that a .PLIT section may not be placed in overlay memory. For more information, see “INPUT_SECTIONS()” on page 3-64.

• OVERLAY_ID() Returns the overlay ID.

• NUMBER_OF_OVERLAYS() Returns the number of overlays that the current link generates when the FIRST_FIT or BEST_FIT overlay placement for ALGO- RITHM() is used. Note: Not currently available.

• ALGORITHM() Directs the linker to use the specified overlay linking algorithm. The only currently available linking algorithm is ALL_FIT.

For ALL_FIT, the linker tries to fit all the OVERLAY_INPUT{} into a single overlay that can overlay into the output section’s run-time memory segment.

(FIRST_FIT – Not currently available.)

VisualDSP++ 5.0 Linker and Utilities Manual 3-71 LDF Keywords, Commands, and Operators

For FIRST_FIT, the linker splits the input sections listed in OVERLAY_INPUT{} into a set of overlays that can each overlay the output section’s run-time memory segment, according to First-In-First-Out (FIFO) order.

(BEST_FIT – Not currently available.) For BEST_FIT, the linker splits the input sections listed in OVERLAY_INPUT{} into a set of overlays that can each overlay the output section’s run-time memory segment, but splits these overlays to optimize memory usage.

• SIZE() Sets an upper limit to the size of the memory that may be occupied by an overlay.

FORCE_CONTIGUITY/NOFORCE_CONTIGUITY

In a section_command, the FORCE_CONTIGUITY command forces contiguous placement of the output section. The NOFORCE_CONTIGUITY command sup- presses a linker warning about non-contiguous placement in the output section.

SHARED_MEMORY{}

The linker can produce two types of executable output—.dxe files and .sm files. A .dxe file runs in a single-processor system’s address space. Shared memory executable (.sm) files reside in the shared memory of a multiprocessor/multi-core system. The SHARED_MEMORY{} command is used to produce .sm files. For more information, see “SHARED_MEMORY{}” on page 5-47.

3-72 VisualDSP++ 5.0 Linker and Utilities Manual 4 EXPERT LINKER

The linker (linker) combines object files into a single executable object module. Using the linker, you can create a new Linker Description File (LDF), modify an existing LDF, and produce an executable file (files). The linker is described in Chapter 2, “Linker”, of this manual. The Expert Linker is a graphical tool that simplifies complex tasks such as memory-mapping manipulation, code and data placement, overlay and shared memory creation, and C stack/heap adjustment. This tool comple- ments the existing VisualDSP++ LDF format by providing a visualization capability enabling new users to take immediate advantage of the powerful LDF format flexibility. Graphics in this chapter demonstrate Expert Linker features. L Some graphics show features not available to all processor families. Processor-specific features are noted in neighboring text. This chapter contains: • “Expert Linker Overview” on page 4-2 • “Launching the Create LDF Wizard” on page 4-3 • “Expert Linker Window Overview” on page 4-9 • “Input Sections Pane” on page 4-10 • “Memory Map Pane” on page 4-16 • “Managing Object Properties” on page 4-47

VisualDSP++ 5.0 Linker and Utilities Manual 4-1 Expert Linker Overview

Expert Linker Overview

Expert Linker is a graphical tool that allows you to: • Define a target processor’s memory map • Place a project’s object sections into that memory map • View how much of the stack or heap has been used after running the DSP program

Expert Linker takes available project information in an .ldf file as input (object files, LDF macros, libraries, and target memory description) and graphically displays it. You can then use drag-and-drop action to arrange the object files in a graphical memory-mapping representation. When you are satisfied with the memory layout, you can generate the executable (.dxe) file via VisualDSP++ project options. Use default LDFs that come with VisualDSP++, or use the Expert L Linker interactive wizard to create new LDFs.

When opening Expert Linker in a project that has an existing .ldf file, Expert Linker parses the .ldf file and graphically displays the target’s memory map and the object mappings. The memory map displays in the Expert Linker window (Figure 4-1). Use this display to modify the memory map or the object mappings. When the project is ready to be built, Expert Linker saves the changes to the .ldf file. Expert Linker is able to show graphically how much space is allocated for program heap and stack. After you load and run the program, Expert Linker can show how much of the heap and stack has been used. You can interactively reduce the amount of space allocated to heap or stack if they are using too much memory. Freeing up memory enables you to store other things like processor code or data.

4-2 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

There are three ways to launch the Expert Linker from VisualDSP++:

• Double-click the .ldf file in the Project window.

• Right-click the .ldf file in the Project window to display a menu and then choose Open in Expert Linker. • From the VisualDSP++ main menu, choose Tools -> Expert Linker -> Create LDF. L This menu item is disabled for Blackfin projects. The Expert Linker window appears.

Figure 4-1. Expert Linker Window

Launching the Create LDF Wizard

The Create LDF Wizard is not available for the Blackfin processor. L Blackfin users should choose Project -> Project Options -> Add Startup Code/LDF to add a new LDF to a project. Also note that Expert Linker cannot be used to modify an LDF file generated by the Project wizard or via the Add Startup Code/LDF page of the Project Options dialog box.

VisualDSP++ 5.0 Linker and Utilities Manual 4-3 Launching the Create LDF Wizard

From the VisualDSP++ main menu, choose Tools -> Expert Linker -> Create LDF to invoke a wizard for creating and customizing a new .ldf file. Use the Create LDF option (Figure 4-1) when creating a new project.

Table 4-1. Welcome Page of the Create LDF Wizard

If an .ldf file is already in the project, you are prompted to confirm whether to create a new .ldf file to replace the existing one. This menu command is disabled when VisualDSP++ does not have a project opened or when the project’s processor-build target is not supported by Expert Linker. Press Next to run the wizard.

4-4 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Step 1: Specifying Project Information

The first wizard window is displayed (Figure 4-2).

Figure 4-2. Selecting File Name and Project Type

You may use or specify the default file name for the .ldf file. The default file name is project_name.ldf, where project_name is the name of the currently opened project. The Project type selection specifies whether the LDF is for a C, C++, assembly, or a VDK project. The default setting depends on the source files in the project. For example, if .c files are in the project, the default is C; if a vdk.h file is in the project, the default is VDK, and so on. This setting determines which template is used as a starting point. For a case where there is a mix of assembly and C files (or any other file combination), the most abstract programming language should be selected. For example, for a project with C and assembly files, a C LDF should be selected. Similarly, for a C++ and C project, the C++ LDF should be selected. Press Next.

VisualDSP++ 5.0 Linker and Utilities Manual 4-5 Launching the Create LDF Wizard

Step 2: Specifying System Information

Choose whether the project is for a single-processor system or a multiprocessor (MP) system (Figure 4-3).

Figure 4-3. Selecting System and Processor Types

By default, the .ldf file is set for single processors. Under System type, select Single processor or Multiprocessor. • For a single-processor system, the Processors list shows only one processor and the MP address columns do not appear. • For a multiprocessor system, right-click in the Processor Properties box to add the desired number of processors included in the .ldf file, name each processor, and set the processor order (which will determine each processor’s MP memory address range). Processor type identifies the DSP system’s processor architecture. This setting is derived from the processor target specified via the Project Options dialog box in VisualDSP++.

4-6 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

By selecting Set up system from debug session settings, the processor information (number of processors and the processor names) is filled automatically from the current settings in the debug session. This field is grayed out when the current debug session is not supported by the Expert Linker. You can also specify the Output file name and the Executables to link against (object libraries, macros, and so on). When you select a processor in the Processors list, the system displays the output file name and the list of executable files to link against for that processor appear. You can change these files by typing a new file name. The file name may include a relative path, an LDF macro, or both. In addition, if the processor’s ID is detected, the processor is placed in the correct position in the processor list. For multiprocessor systems, the window (Figure 4-4) shows the list of processors in the project.

Figure 4-4. Processors and MMS Offset

VisualDSP++ 5.0 Linker and Utilities Manual 4-7 Launching the Create LDF Wizard

Expert Linker automatically displays MP address range for each processor space providing specific MP addresses and multiprocessor memory space (MMS) offsets which makes using MP commands much easier. This is an automatic replacement for the MPMEMORY linker command used in the .ldf source file. The MP address range is available only for processors that have MP L memory space. Press Next to advance to the Wizard Completed page.

Step 3: Completing the LDF Wizard

From the Wizard Completed page, you can go back and verify or modify selections made up to this point.

When you click the Finish button, Expert Linker copies a template .ldf file to the same directory that contains the project file and adds it to the current project. The Expert Linker window appears and displays the contents of the new .ldf file.

Figure 4-5. Wizard Completed Page of the Create LDF Wizard

4-8 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Expert Linker Window Overview

The Expert Linker window contains two panes: • The Input Sections pane (Figure 4-6) provides a tree display of the project’s input sections (see “Input Sections Pane” on page 4-10). • The Memory Map pane displays each memory map in a tree or graphical representation (see “Memory Map Pane” on page 4-16).

Figure 4-6. Expert Linker Window

Using LDF commands, the linker reads the input sections from object (.doj) files and places them in output sections in the executable file. The LDF defines the processor’s memory and indicates where within that memory the linker is to place the input sections. Using drag-and-drop, you can map an input section to an output section in the memory map. Each memory segment may have one or more output sections under it. Input sections that have been mapped to an output section are displayed under that output section. For more information, refer to “Input Sections Pane” on page 4-10 and “Memory Map Pane” on page 4-16.

VisualDSP++ 5.0 Linker and Utilities Manual 4-9 Input Sections Pane

Access various Expert Linker functions with your mouse. L Right-click to display appropriate menus and make selections.

Input Sections Pane

The Input Sections pane initially displays a list of all the input sections referenced by the .ldf file, and all input sections contained in the object files and libraries. Under each input section, a list of LDF macros, libraries, and object files may be contained in that input section. You can add or delete input sections, LDF macros, or objects/library files in this pane.

Input Sections Menu

Right-click an object in the Input Sections pane, and a menu appears as shown in Figure 4-7.

Figure 4-7. Input Sections Right-Click Menu

4-10 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

The main menu functions include: • Sort by – Sorts objects by input sections or LDF macros. These selections are mutually exclusive. • Add – Adds input sections, object/library files, and LDF macros. Appropriate menu selections are grayed out when right-clicking on a position (area) in which you cannot create a corresponding object. Create an input section as a shell, without object/library files or LDF macros in it. You can even map this section to an output section. However, input sections without data are grayed out. • Delete – Deletes the selected object (input section, object/library file, or LDF macro). • Remove – Removes an LDF macro from another LDF macro but does not delete the input section mappings that contain the removed macro. The difference between Delete and Remove is that Delete completely deletes the input section macros that contain the deleted macro.

NOTE: The Remove option becomes available only if you right-click on an LDF macro that is part of another LDF macro. • Expand All LDF Macros – Expands all the LDF macros in the input sections pane so that the contents of all the LDF macros are visible. • View Legend – Displays the Legend dialog box which shows icons and colors used by the Expert Linker.

VisualDSP++ 5.0 Linker and Utilities Manual 4-11 Input Sections Pane

• View Section Contents – Opens the Section Contents dialog box, which displays the section contents of the object file, library file, or .dxe file. This command is available only after you link or build the project and then right-click on an object or output section. • View Global Properties – Displays the Global Properties dialog box which provides the map file name (of the map file generated after linking the project) as well as access to various processor and setup information (see Figure 4-42 on page 4-48).

Mapping an Input Section to an Output Section

Using the Expert Linker, you can map an input section to an output section. By using Windows drag-and-drop action, click on the input section, drag the mouse pointer to an output section, and then release the mouse button to drop the input section onto the output section. All objects, such as LDF macros or object files under that input section, are mapped to the output section. Once an input section has been mapped, the icon next to the input section changes to denote that it is mapped. If an input section is dragged onto a memory segment with no output section in it, an output section with a default name is automatically created and displayed. A red “x” on an icon indicates the object/file is not mapped. Once an input section has been completely mapped (that is, all object files that contain the section are mapped), the icon next to the input section changes to indicate that it is now mapped; the “x” disappears. See Figure 4-8. As you drag the input section, the icon changes to a circle with a diagonal slash if it is over an object where you are not allowed to drop the input section.

4-12 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Viewing Icons and Colors

Use the Legend dialog box to display all possible icons in the tree pane as well as short descriptions of each icon. (Figure 4-8)

Figure 4-8. Legend Dialog Box – Icons Page L The red “x” on an icon indicates this object/file is not mapped. Click the Colors tab to view the Colors page (Figure 4-9). This page contains a list of colors used in the graphical memory map view; each item’s

VisualDSP++ 5.0 Linker and Utilities Manual 4-13 Input Sections Pane

color can be customized. The list of displayed objects depends on the processor family.

Figure 4-9. Legend Dialog Box – Colors Page

To change a color: 1. Double-click the color. You can also right-click on a color and select Properties. The system displays the Select a Color dialog box (Figure 4-10). 2. Select a color and click OK.

Click Other to select other colors from the advanced palette. Click Reset to reset all memory map colors to the default colors.

Figure 4-10. Select a Color Dialog Box

4-14 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Sorting Objects

Objects in the Input Sections pane can be sorted by input sections (default) or by LDF macros, like $OBJECTS or $COMMAND_LINE_OBJECTS. The Input Sections and LDF Macros menu selections are mutually exclusive—only one can be selected at a time. Refer to Figure 4-11 and Figure 4-12.

Figure 4-11. Expert Linker Window – Sorted by Input Sections

Figure 4-12. Expert Linker Window – Sorted by LDF Macros

Other macros, object files, or libraries may appear under each macro. Under each object file are input sections contained in that object file. When the tree is sorted by LDF macros, only input sections can be L dragged onto output sections.

VisualDSP++ 5.0 Linker and Utilities Manual 4-15 Memory Map Pane

Memory Map Pane

In an .ldf file, the linker’s MEMORY() command defines the target system’s physical memory. Its argument list partitions memory into memory segments and specifies start and end addresses, memory width, and memory type (such as program, data, stack, and so on). It connects your program to the target system. The OUTPUT() command directs the linker to produce an executable (.dxe) file and specifies its file name. Figure 4-13 shows a typical memory map pane.

Processor Tab

Figure 4-13. Expert Linker Window – Memory Map

This section describes: • “Context Menu” on page 4-19 • “Tree View Memory Map Representation” on page 4-21 • “Graphical View Memory Map Representation” on page 4-22 • “Specifying Pre- and Post-Link Memory Map View” on page 4-26 • “Zooming In and Out on the Memory Map” on page 4-28

4-16 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

• “Adding a Memory Segment” on page 4-29 • “Inserting a Gap Into a Memory Segment” on page 4-31 • “Working With Overlays” on page 4-32 • “Viewing Section Contents” on page 4-33 • “Viewing Symbols” on page 4-36 • “Profiling Object Sections” on page 4-37 • “Adding Shared Memory Segments and Linking Object Files” on page 4-42 The Memory Map pane has tabbed pages. You can page through the memory maps of the processors and shared memories to view their makeup. The two viewing modes are a tree view and a graphical view. Select these views and other memory map features by means of the right-click (context) menu. All procedures involving memory map handling assume the Expert Linker window is open. The Memory Map pane displays a tooltip when the mouse cursor moves over an object in the display. The tooltip shows the object’s name, address, and size. The system also uses representations of overlays, which display in “run” space and “live” space. Use the right-click menu (“Context Menu” on page 4-19) to select and perform major memory map functions. Invalid Memory Segment Notification: When a memory segment is invalid (for example, when a memory range overlaps another memory segment or if the memory width is invalid), the tree shows an Invalid Memory Segment icon (see Figure 4-14). Move the

VisualDSP++ 5.0 Linker and Utilities Manual 4-17 Memory Map Pane

mouse pointer over the icon and a tooltip displays a message describing why the segment is invalid.

Invalid Memory Segments

Figure 4-14. Memory Map With Invalid Memory Segments

4-18 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Context Menu

Display the context menu by right-clicking in the Memory Map pane. This menu (Figure 4-15) allows you to select and perform major functions. The available right-click menu commands are listed below.

Figure 4-15. Memory Map Main Menu

View Mode • Memory Map Tree – Displays the memory map in a tree representation (see Figure 4-16 on page 4-21) • Graphical Memory Map – Displays the memory map in graphical blocks (see Figure 4-17 on page 4-23) View • Mapping Strategy (Pre-Link) – Displays the memory map that shows the placement of your object sections. • Link Results (Post-Link) – Displays the memory map that shows the actual placement of the object sections.

VisualDSP++ 5.0 Linker and Utilities Manual 4-19 Memory Map Pane

New • Memory Segment – Specifies the name, address range, type, size, and so on for memory segments you want to add. • Output Section – Adds an output section to the selected memory segment. (Right-click on the memory segment to access this command.) If you do not right-click on a memory segment, this option is disabled. • Shared Memory – Adds a shared memory to the memory map. • Overlay – Invokes a dialog box that allows adding a new overlay to the selected output section or memory segment. The selected output section is the new overlay’s run space (see Figure 4-53 on page 4-65). Delete – Deletes the selected object. Expand All – Expands all items in the memory map tree so that their contents are visible. Pin to Output Section – Pins an object section to an output section to prevent it from overflowing to another output section. This command appears only when right-clicking an object section that is part of an output section specified to overflow to another output section. View Section Contents – Invokes a dialog box that displays the contents of the input or output section. It is available only after you link or build the project and then right-click on an input or object section (see Figure 4-30 on page 4-35). View Symbols – Invokes a dialog box that displays the symbols for the project, overlay, or input section. It is available only after you link the project and then right-click on a processor, overlay, or input section (see Figure 4-42 on page 4-48).

4-20 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Properties – Displays a Properties dialog box for the selected object. The Properties menu is context-sensitive; different properties are displayed for different objects. Right-click a memory segment and choose Properties to specify a memory segment’s attributes (name, start address, end address, size, width, memory space, PM/DM/(BM), RAM/ROM, and internal or external flag). View Legend – Displays the Legend dialog box showing tree view icons and a short description for each icon. The Colors page lists the colors used in the graphical memory map. You can customize each object’s color. See Figure 4-8 on page 4-13 and Figure 4-9 on page 4-14. View Global Properties – Displays a Global Properties dialog box that lists the map file generated after linking the project. It also provides access to some processor and setup information (see Figure 4-43 on page 4-49).

Tree View Memory Map Representation

In the tree view (selected by right-clicking and choosing View Mode -> Memory Map Tree), the memory map is displayed with memory segments at the top level (Figure 4-16 on page 4-21).

Processor Tab

Figure 4-16. Expert Linker Window – Memory Map Tree View

VisualDSP++ 5.0 Linker and Utilities Manual 4-21 Memory Map Pane

Each memory segment may have one or more output sections under it. Input sections mapped to an output section appear under that output section. The start address and size of the memory segment display in separate columns. If available, the start address and the size of each output section are displayed (for example, after you link the project).

Graphical View Memory Map Representation

In the graphical view (selected by right-clicking in the Memory Map pane and choosing View Mode -> Graphical Memory Map), the graphical memory map (Figure 4-17) displays the processor’s hardware memory map (refer to your processor’s hardware reference manual or data sheet). Each hardware memory segment contains a list of user-defined memory segments. View the memory map from two perspectives: pre-link view and post-link view (see “Specifying Pre- and Post-Link Memory Map View” on

4-22 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

page 4-26). Figure 4-17 through Figure 4-21 show examples of graphical memory map representations.

Output Sections

Intput Sections

User-defined Memory Segments

Figure 4-17. Graphical Memory Map Representation

In graphical view, the memory map comprises blocks of different colors that represent memory segments, output sections, objects, and so on. The memory map is drawn with these rules: • An output section is represented as a vertical header with a group of objects to the right of it. • A memory segment’s border and text change to red (from its normal black color) to indicate that it is invalid. When moving the mouse pointer over the invalid memory segment, a tooltip displays a message, describing why the segment is invalid.

VisualDSP++ 5.0 Linker and Utilities Manual 4-23 Memory Map Pane

• The height of the memory segments is not scaled as a percentage of the total memory space. However, the width of the memory segments is scaled as a percentage of the widest memory. • Object sections are drawn as horizontal blocks stacked on top of each other. Before linking, the object section sizes are not known and are displayed in equal sizes within the memory segment. After linking, the height of the objects is scaled as a percentage of the total memory segment size. Object section names appear only when there is enough room to display them. • Addresses are listed in ascending order from top to bottom. Three buttons at the top right of the Memory Map pane permit zooming. If there is not enough room to display the memory map when zoomed in, horizontal and/or vertical scroll bars allow you to view the entire memory map (for more information, see “Zooming In and Out on the Memory Map” on page 4-28). You can drag-and-drop any object except memory segments. See Figure 4-18.

1. Select an input section

2. Drag it to an output section

Figure 4-18. Dragging and Dropping an Object (1)

4-24 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

1. Select an output section

2. Drag it to a new location

Figure 4-19. Dragging and Dropping an Object (2)

Select a memory segment to display its border. Memory segments, when selected, display a tiny box at their top and bottom borders (Figure 4-20). Drag the border (at this box) to change the memory segment’s size. By doing this, the size of the selected and adjacent memory segments change.

Move this box to adjust the size of the memory segment

Figure 4-20. Adjusting the Size of a Memory Segment

VisualDSP++ 5.0 Linker and Utilities Manual 4-25 Memory Map Pane

When the mouse pointer is on top of the box, the resize cursor appears as

When an object is selected in the memory map, it is highlighted as L shown in Figure 4-21 on page 4-26. If you move the mouse pointer over an object in the graphical memory map, a yellow tooltip displays the information about the object (such as name, address, and size).

Highlighted Object

Figure 4-21. A Highlighted Memory Segment in the Memory Map

Specifying Pre- and Post-Link Memory Map View

View the memory map from two perspectives: pre-link view and post-link view. Pre-link view is typically used to place input sections. Post-link view is typically used to view where the input sections are placed after linking

4-26 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

the project. Other information (such as the sizes of each section, symbols, and the contents of each section) is available after linking. • To enable pre-link view from the Memory Map pane, right-click and choose View and Mapping Strategy (Pre-Link). Figure 4-22 on page 4-27 illustrates a memory map before linking.

Figure 4-22. Memory Map Pane in Pre-Link View

• To enable post-link view from the Memory Map pane, right-click and choose View and Link Results (Post-Link). Figure 4-23 on

VisualDSP++ 5.0 Linker and Utilities Manual 4-27 Memory Map Pane

page 4-28 illustrates a memory map after linking.

Figure 4-23. Memory Map Pane in Post-Link View

Zooming In and Out on the Memory Map

From the Memory Map pane, you can zoom in or out incrementally or zoom in or out completely. Three buttons at the top right of the pane perform zooming operations. Horizontal and/or vertical scroll bars appear when there is not enough room to display a zoomed memory map in the Memory Map pane (see Figure 4-24 on page 4-28).

Zoom Options

Figure 4-24. Memory Map – Zoom Options

4-28 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

To: • Zoom in, click on the magnifying glass icon with the + sign above the upper right corner of the memory map window. • Zoom out, click on the magnifying glass icon with the - sign above the upper right corner of the memory map window. • Exit zoom mode, click on the magnifying glass icon with the “x” above the upper right corner of the memory map window. • View a memory object by itself by double-clicking on the memory object. • View the memory object containing the current memory object by double-clicking on the white space around the memory object

Adding a Memory Segment

You can add memory segments to the memory map. This procedure assumes that the Expert Linker window (Memory Map pane) is open. To add a memory segment: 1. Right-click in the Memory Map pane. 2. Choose New and then choose Memory Segment. The Memory Segment Properties dialog box appears (Figure 4-25 on page 4-30). 3. In Name, type a name for the memory segment.

VisualDSP++ 5.0 Linker and Utilities Manual 4-29 Memory Map Pane

4. Specify the following attributes:

Figure 4-25. Memory Segment Properties Dialog Box

• Start address • End address • Size (hexadecimal) It is only necessary to specify either “End Address” or “Length” and L not both. • Width • Memory Space For Blackfin and TigerSHARC processors, this option is unavailable, because VisualDSP++ employs a unified memory space. • RAM/ROM/SROM

4-30 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

• Internal/External (memory location) 5. Click OK.

Inserting a Gap Into a Memory Segment

A gap may be inserted into a memory segment in the graphical memory map. To insert a gap: 1. Right-click on a memory segment. 2. Choose Insert gap. The Insert Gap dialog box appears, as shown in Figure 4-26. It displays the start address, end address, and size of the selected memory segment.

Figure 4-26. Insert Gap Dialog Box

VisualDSP++ 5.0 Linker and Utilities Manual 4-31 Memory Map Pane

You may insert a gap at the start of the memory segment or the end of it. •If the Start... is chosen, the Start address for the gap is grayed out and you must enter an End Address or Size (of the gap). •If the End... is chosen, the End address of the gap is grayed out and you must enter a Start Address or Size.

Working With Overlays

Overlays appear in the memory map window in two places: “run” space and “live” space. Live space is where the overlay is stored until it is swapped into run space. Because multiple overlays can exist in the same “run” space, the overlays display as multiple blocks on top of each other in cascading fashion. Figure 4-27 shows an overlay in “live” space, and Figure 4-28 shows an overlay in “run” space.

Figure 4-27. Graphical Memory Map Showing an Overlay in “Live” Space

Overlays in a “run” space appear one at a time in the graphical memory map. The scroll bar next to an overlay in “run” space allows you to specify an overlay to be shown on top. Drag the overlay on top to another output section to change the “run” space for an overlay.

4-32 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Click the Up arrow or Down arrow button in the header to display a previous overlay or next overlay in “run” space. Click the browse button to display the list of all available overlays. The header shows the number of overlays in this “run” space as well as the current overlay number.

Browse Button Figure 4-28. Graphical Memory Map Showing an Overlay “Run” Space

To create an overlay in the “run” space: 1. Right-click on an output section. 2. Choose New -> Overlay. 3. Select the “live” space from the Overlay Properties dialog box (see “Managing Overlay Properties” on page 4-65). The new overlay appears in the “run” and “live” spaces in two different colors in the memory map.

Viewing Section Contents

To view the contents of an input section or an output section, specify the particular memory address and the display’s format.

VisualDSP++ 5.0 Linker and Utilities Manual 4-33 Memory Map Pane

This capability employs the elfdump utility (elfdump.exe) to obtain the section contents and display it in a window similar to a memory window in VisualDSP++. Multiple Section Contents dialog boxes may be displayed. For example, Figure 4-29 shows output section contents in HEX format.

Figure 4-29. Output Section Contents in Hex Format

4-34 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

To display the contents of an output section: 1. In the Memory Map pane, right-click an output section. 2. Choose View Section Contents from the menu. The Section Contents dialog box appears. By default, the memory section content appears in Hex format. 3. Right-click anywhere in the section view to display a menu with these selections: • Go To – Displays an address in the window. • Select Format — Provides a list of formats: Hex, Hex and ASCII, and Hex and Assembly. Select a format type to specify the memory format. Figure 4-30 and Figure 4-31 illustrate memory data formats available for the selected output section.

Figure 4-30. Output Section Contents in Hex and ASCII Format

VisualDSP++ 5.0 Linker and Utilities Manual 4-35 Memory Map Pane

Figure 4-31. Output Section Contents in Hex and Assembly Format

Viewing Symbols

Symbols can be displayed per a processor program (.dxe), per overlay (.ovl), or per input section. Initially, symbol data is in the same order in which it appears in the linker’s map output. Sort symbols by name, address, and so on by clicking the column headings.

4-36 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

To view symbols (Figure 4-32):

Figure 4-32. View Symbols Dialog Box

1. In the post-link view of the Memory Map pane, select the item (memory segment, output section, or input section) whose symbols you want to view. 2. Right-click and choose View Symbols. The View Symbols dialog box displays the selected item’s symbols. The symbol’s address, size, binding, file name, and section appear beside the symbol’s name.

Profiling Object Sections

Use Expert Linker to profile object sections in your program. After doing so, Expert Linker graphically displays how much time was spent in each object section so you can locate code “hotspots” and move the code to faster, internal memory. The following is a sample profiling procedure.

VisualDSP++ 5.0 Linker and Utilities Manual 4-37 Memory Map Pane

Start by selecting Profile execution of object sections in the General page of the Global Properties dialog box (Figure 4-33).

Figure 4-33. General Page of the Global Properties Dialog Box

Then build the project and load the program. After the program is loaded, Expert Linker sets up the profiling bins to collect the profiling information.

4-38 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

When the program run is complete, Expert Linker colors each object section with a different shade of red to indicate how much time was spent executing that section. For an example, see Figure 4-34.

Figure 4-34. Colored Object Sections

The fir.doj (seg_pmco) section appears in the brightest shade of red, indicating that it takes up most of the execution time. The shading of the libio.dlb (seg_pmco) section is not as bright. This indicates that it takes up less execution time than fir.doj (seg_pmco). The shading of the libc.dlb (seg_pmco) section is black, indicating that it takes up a negligi- ble amount of the total execution time.

VisualDSP++ 5.0 Linker and Utilities Manual 4-39 Memory Map Pane

From Expert Linker, you can view PC sample counts for object sections. To view an actual PC sample count (Figure 4-35), move the mouse pointer over an object section and view the PC sample count.

Figure 4-35. PC Sample Count

To view sample counts for functions located within an object section, double-click on the object section (Figure 4-36).

Figure 4-36. Sample Count of Functions Within Object Section

Functions are available only when objects are compiled with debug L information.

4-40 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

You can view detailed profile information such as the sample counts for each line in the function (Figure 4-37). To view detailed profile information, double-click on a function.

Figure 4-37. Profile Information (Detail)

To view PC samples as a percentage of total samples, view the memory map tree (Figure 4-38).

Figure 4-38. Percentage of Total PC Sample Count

VisualDSP++ 5.0 Linker and Utilities Manual 4-41 Memory Map Pane

Adding Shared Memory Segments and Linking Object Files

In many DSP applications where large amounts of memory for multiprocessing tasks and sharing of data are required, an external resource in the form of shared memory may be desired. Refer to Engineer-to- Engineer Note EE-202 “Using the Expert L Linker for Multiprocessor LDF” for a detailed description and procedure. Find this EE Note on Analog Devices Web site at: http://www.analog.com/ee-notes.

Figure 4-39. Multiprocessor LDF Selection

To add a shared memory section to the .LDF file, right-click in the Memory Map pane and select New/Shared Memory. Then specify a name for the shared memory segment (.sm) and select the processors that have access to this shared memory segment. Refer to “Managing Shared Mem- ory Properties” on page 4-70 for more information.

4-42 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

As shown in Figure 4-40, a new shared memory segment, visible to processors P0 and P1, has been successfully added to the system. Note that variables declared in the shared memory segment will be accessed by both processors in the system. In order for the linker to be able to correctly resolve these variables, the link against command should be used once again.

Figure 4-40. Shared Memory Segment

Expert Linker automatically adds shared memory segments, and therefore no any additional modifications to the LDF are needed.

Confirm that Expert Linker has correctly added the .sm file to the link against command line by selecting View Global Properties in the Memory Map pane and clicking on the Processor tab.

VisualDSP++ 5.0 Linker and Utilities Manual 4-43 Memory Map Pane

The shared.sm file should now be contained in the Executables to Link Against box for each processor. Use Expert Linker to detect non-linked input sections, such as a variable declared in external SDRAM memory, which belongs to the shared memory segment. When both processors and the shared memory segments have been properly configured, and Expert Linker has detected all input sections, you can link the object files from different input sections to their corresponding memory sections. In general, the linking process consists of these steps: 1. Sort the left pane of the Expert Linker window by LDF macros instead of input sections (default setting). To do that, right-click on the left pane and select Sort by/LDF Macros. 2. Right-click on the LDF Macro window and add a new macro for P0 (Add/LDF Macro). For example, $OBJECTS_P0. Repeat the same step for P1 and shared.sm.

3. Add the object (.doj) files that correspond to each processor as well as to the shared memory segment. To do this, right-click on each recently created LDF macro and then select Add/Object/Library File. The use of LDF macros becomes extremely useful in systems where there is more than one object files, .doj files per processor or shared memory segments, in which case the same step previously explained should be followed for each .doj file.

4. Delete the LDF macro, $COMMAND_LINE_OBJECTS, from the $OBJECTS macro to avoid duplicate object files during the linking process. Right-click on the $COMMAND_LINE_OBJECTS macro and click Remove.

4-44 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

5. The left pane needs to be sorted by Input Sections instead of LDF macros. To do that, right-click on the left pane and select Sort by/Input Sections. Additionally, in the right pane, change the Memory Map View Mode from Graphical to Tree mode. Right-click on the Memory Map window, select View Mode, and then Memory Map Tree. 6. Map the new macros into memory. To do this, place each macro into its corresponding memory section.

7. Repeat the same steps for processor P1 ($OBJECTS_P1) and for the shared memory segment, shared.sm (place $OBJECTS_SM in the SDRAM section). 8. Press Rebuild All. 9. Select one of the processors by clicking on the processor’s name tab. In this case, P0 is selected first. Then, place (drag-and-drop) the recently created LDF macro, $OBJECTS_P0, in its corresponding memory segment. The red crosses denoting the “non-linked” sections have disappeared, indicating that the input sections have been properly mapped into memory. Also, note that the LDF macros that were moved from the Input L Sections window (left pane) to their corresponding sections in the Memory Map window (right pane) have been automatically replaced during the linking process with the actual object files used by the linker.

VisualDSP++ 5.0 Linker and Utilities Manual 4-45 Memory Map Pane

The LDF is now complete. Figure 4-41 illustrates the generated LDF in the Source Code View mode.

Figure 4-41. Expert Linker Multiprocessor LDF

The multiprocessor linker commands, MPMEMORY, SHARED MEMORY and LINK AGAINST, as well as the corresponding LDF macros, were successfully generated by the Expert Linker in a way absolutely transparent to the user. The complete project is now ready to be built. Once again, perform a Rebuild All and start debugging with the application code.

4-46 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Managing Object Properties

You can display different properties for each type of object. Since different objects may share certain properties, their Properties dialog boxes share pages. The following procedures assume the Expert Linker window is L open. To display a Properties dialog box, right-click an object and choose Properties. You may choose these functions: • “Managing General Global Properties” on page 4-48 • “Managing Processor Properties” on page 4-49 • “Managing PLIT Properties for Overlays” on page 4-50 • “Managing Elimination Properties” on page 4-51 • “Managing Symbols Properties” on page 4-53 • “Managing Memory Segment Properties” on page 4-57 • “Managing Output Section Properties” on page 4-58 • “Managing Packing Properties” on page 4-61 • “Managing Alignment and Fill Properties” on page 4-63 • “Managing Overlay Properties” on page 4-65 • “Managing Stack and Heap in Processor Memory” on page 4-67 • “Managing Shared Memory Properties” on page 4-70

VisualDSP++ 5.0 Linker and Utilities Manual 4-47 Managing Object Properties

Managing General Global Properties

To access Global Properties, right-click in the Input Sections pane and choose Properties. The Global Properties dialog box appears. The General tab of the Global Properties dialog box provides these selections (Figure 4-42): • Linker map file displays the map file generated after linking the project. This is a read-only field. •If Show stack/heap usage is selected after you run a project, Expert Linker shows how much of the stack and heap were used. •If Profile execution of object sections is selected, Expert Linker enables the profiling feature that allows you to see “hotspots” in object sections and to fine-tune the placement of object sections.

Figure 4-42. General Page of the Global Properties Dialog Box

4-48 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Managing Processor Properties

To specify processor properties: 1. In the Memory Map pane, right-click on a Processor tab and choose Properties. The Processor Properties dialog box appears. 2. Click the Processor tab (Figure 4-43). The Processor tab allows you to reconfigure the processor setup.

Figure 4-43. Processor Page of the Processor Properties Dialog Box

With a Processor tab in focus, you can: • Specify System Type – It may be a Single processor or Multipro- cessor selection. (The Processors list displays the names of all the processors in the project and the address range for each processor.)

• Select a Processor type (such as ADSP-21060).

VisualDSP++ 5.0 Linker and Utilities Manual 4-49 Managing Object Properties

• Specify an Output file name – The file name may include a relative path and/or LDF macro. Specify an output file for each processor. • Specify Executables to link against – Multiple files names are permitted, but must be separated with space characters or commas. Only .sm, .dlb, and .dxe files are permitted. A file name may include a relative path, LDF macro, or both. Additionally, a processor can be renamed by selecting the processor, right-clicking, choosing Rename Processor, and typing a new name. For multiprocessor systems, you can add, delete, and rearrange processor order. Right-click in the Processors box, choose Add Processor, and type a name for the new processor, or choose Delete Processor. To move a processor, select the processor and drag it to another position in the Processors list. When a processor in a multiprocessor system is moved to a differ- L ent position, its address range changes. The MP Start Addr. and MP End Addr. information is static.

Managing PLIT Properties for Overlays

The PLIT tab allows you to view and edit the function template used in overlays. Assembly instructions observe the same syntax coloring as specified for editor windows. L Enter assembly code only. Comments are not allowed.

4-50 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

To view and edit PLIT information: 1. Right-click in the Input Sections pane. 2. Choose Properties. The Global Properties dialog box appears. 3. Click the PLIT tab (Figure 4-44).

Figure 4-44. PLIT Page of the Global Properties Dialog Box

Managing Elimination Properties

Eliminate unused code from the target .dxe file. Specify the input sections from which to eliminate code and the symbols you want to keep. Select the Global Properties dialog box by right-clicking in the Input Sections pane and choosing Properties.

VisualDSP++ 5.0 Linker and Utilities Manual 4-51 Managing Object Properties

Use the Elimination tab to perform elimination (Figure 4-45).

Figure 4-45. Elimination Tab

Selecting the Enable elimination of unused objects option enables elimination. This check box is grayed out when elimination is enabled through the linker command line or when the .ldf file is read-only. When Verbose linker output of eliminated objects is selected, the eliminated objects are shown as linker output in the Output window’s Build page during linking. This check box is grayed out when the Enable elimination of unused objects check box is cleared. It is also grayed out when elimination is enabled through the linker command line or when the .ldf file is read-only. The Sections to apply elimination box lists all input sections with a check box next to each section. Elimination applies to the sections that are selected. By default, all input sections are selected. The Symbols to keep box displays a list of symbols to be retained (see “Managing Symbols Properties” on page 4-53 for more information).

4-52 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Managing Symbols Properties

You can view the list of symbols resolved by the linker. You can also add and remove symbols from the list of symbols kept by the linker. The symbols can be resolved to an absolute address or to a program (.dxe) file. It is assumed that the elimination of unused code is enabled. To add or remove a symbol: 1. Right-click in the Input Sections pane. 2. Choose Properties. The Global Properties dialog box appears.

VisualDSP++ 5.0 Linker and Utilities Manual 4-53 Managing Object Properties

3. Click the Elimination tab to add or remove a symbol (Figure 4-46).

Figure 4-46. Elimination Page of he Global Properties Dialog Box

4. Right-click in the Symbols to keep box. Using the menu, choose Add Symbol to open the dialog box and type a new symbol name (names) at the end of the existing list. To delete a symbol, select the symbol, right-click, and choose Remove Symbol. To specify symbol resolution: 1. In the Memory Map pane, right-click a Processor tab. 2. Choose Properties. The Processor page of the Processor Properties dialog box appears. The Symbols tab allows you to specify how symbols are to be

4-54 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

resolved by the linker (Figure 4-47).

Figure 4-47. Processor Properties Dialog Box – Symbols Tab

The symbols can be resolved to an absolute address or to a program file. Right-clicking in the Symbols field allows you to add or remove symbols. Choosing Add Symbol from the menu invokes the Add Symbol to Resolve dialog box (Figure 4-48), which allows you to pick a symbol by either typing the name or browsing for a symbol. Using Resolve with, you

VisualDSP++ 5.0 Linker and Utilities Manual 4-55 Managing Object Properties

can also decide whether to resolve the symbol from a known absolute address or file name (.dxe or .sm) file.

Figure 4-48. Add Symbol to Resolve Dialog Box

The Browse button is grayed out when no symbol list is available; for example, if the project has not been linked. When this button is active, click it to display the Browse Symbols dialog box, which shows a list of all the symbols. Selecting a symbol from that list places it in the Symbol box of the Edit Symbol to Resolve dialog box. To delete a symbol from the resolve list: 1. Click Browse to display the Symbols to resolve list (Figure 4-48). 2. Select the symbol to delete. 3. Right-click and choose Remove Symbol.

4-56 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Managing Memory Segment Properties

Specify or change the memory segment’s name, start address, end address, size, width, memory space, memory type, and internal/external flag. To display the Memory Segment Properties dialog box (Figure 4-49 on page 4-57):

1. Right-click a memory segment (for example, PROGRAM or MEM_CODE) in the Memory Map pane. 2. Choose Properties. The selected segment properties are displayed.

Figure 4-49. Memory Segment Properties Dialog Box

VisualDSP++ 5.0 Linker and Utilities Manual 4-57 Managing Object Properties

Managing Output Section Properties

Use the Output Section tab to change the output section’s name or to set the overflow (Figure 4-50).

Figure 4-50. Output Section Properties Dialog Box – Output Section Tab

Overflow allows objects that do not fit in the current output section to spill over into the specified output section. By default, all objects that do not fit (except objects that are manually pinned to the current output section) overflow to the specified section. To specify output section properties:

1. Right-click an output section (for example, PROGRAM_DXE or CODE_DXE) in the Memory Map pane. 2. Choose Properties. The selections in the output section/segment list include “None” (for no overflow) and “All” output sections. Pin objects to an output section by right-clicking the object and choosing Pin to output section.

4-58 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

You can: • Type a name for the output section in Name. •In Overflow, select an output section into which the selected output section will overflow; select None for no overflow. This setting appears in the Placement box. Before linking the project, the Placement box indicates the output section’s address and size as “Not available”. After linking is done, the box displays the output section’s actual address and size. • Initialization allows you to choose the initialization qualifier for an output section. The section qualifier set by this option controls the operation of run-time initialization by tools that process the executable file and the run-time initialization that can be achieved by enabling the meminit utility. The choices are: • None: Stipulates no special treatment for the section – the section data are statically initialized according to their definition in the source, no runtime initialization is called for. Please note that data that have no explicit initialization in source are initialized to 0. • No initialization: Stipulates no data initialization, even statically. No data for the section are in the executable file. This is equivalent to specifying a section qualifier SHT_NOBITS in the LDF. • Initialize to zero: The memory space for this section will be initialized to zero at either “load” or “runtime”, if invoked with the linker’s -meminit switch. If the -meminit switch is not used, the memory is initialized at “load” time when the .dxe file is loaded via VisualDSP++ IDDE, or boot-loaded

VisualDSP++ 5.0 Linker and Utilities Manual 4-59 Managing Object Properties

by the boot kernel. If the memory initializer is invoked, the C/C++ run-time library (CRTL) will process embedded information to initialize the memory space during the CRTL initialization process.

• Initialize at runtime: If the linker is invoked with the -meminit switch, this section will be filled at runtime. If the -meminit switch is not specified, the section is filled at “load” time. • Contiguity of Input Sections allows you to choose whether or not code or data in an output section should be mapped contiguously. The choices are: • Display linker warning if section is not mapped contiguously • Force contiguous placement of sections • Suppress linker warning about non-contiguous placement of sections in the operating system • Specify the Packing (on page 4-61) and Alignment (with Fill value) properties (on page 4-63) as needed.

4-60 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Managing Packing Properties

Use the Packing tab to specify the packing format that the linker employs to place bytes into memory. The choices include No packing or Custom packing. You can view byte order, which defines the order that bytes will be placed into memory, and you can change this order. It can be viewed via the Packing order box. To specify packing properties: 1. Right-click a memory segment in the Memory Map pane. 2. Choose Properties and click the Packing tab (Figure 4-51).

Figure 4-51. Memory Segment Properties Dialog Box – Packing Tab

VisualDSP++ 5.0 Linker and Utilities Manual 4-61 Managing Object Properties

3. In Packing method, select a method.

Method Description

No packing Specifies no packing. Number of bytes and Packing order are grayed out.

Custom Permits the selection of number of bytes and packing order. Other choices Specifies the number of bytes and packing order of the selected method. The list of packing methods is derived from the included packing.h file. Packing method information (number of bytes and packing order) appears, but you cannot change it.

4. In Number of bytes (if Custom is selected), specify the number of bytes to be reordered at one time. This value does not include the number of null bytes inserted into memory. 5. In Packing order, specify byte packing. To do that, select a byte and perform one of these actions: • Click the keyboard's Up arrow or Down arrow key. • Drag and drop it to a new location. • Insert a null byte by clicking on Insert. • Delete a null byte by selecting the null byte and clicking Delete. 6. Click OK.

4-62 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Managing Alignment and Fill Properties

Use the Alignment tab to set the alignment and fill values for the output section. When the output section is aligned on an address, the linker fills the gap with zeros (0), NOP instructions, or a specified value. To specify alignment properties: 1. Right-click a memory segment in the Memory Map pane. 2. Choose Properties. 3. Click the Alignment tab (Figure 4-52).

Figure 4-52. Output Section Properties – Alignment Tab

If you select No Alignment, the output section is not be aligned on an address. If you choose Align each input section to the next address that is a multiple of, select an integer value from the drop-down list to specify the output section alignment.

VisualDSP++ 5.0 Linker and Utilities Manual 4-63 Managing Object Properties

When the output section is aligned on an address, a gap is filled by the linker. Based on the processor architecture, Expert Linker determines the opcode for the NOP instruction.

The Fill value is either 0 (default), a NOP instruction, or a user-specified value (a hexadecimal value entered in the entry box).

4-64 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Managing Overlay Properties

Use the Overlay tab to add/choose the output file for the overlay, its “live” memory, and its linking algorithm. To specify overlay properties: 1. Right-click an overlay object in the Memory Map pane. 2. Choose Properties and click the Overlay tab (Figure 4-53)

Figure 4-53. Overlay Properties Dialog Box – Overlay Tab

Use the Output file name box to specify the name of the overlay file (.ovl). The Live Memory drop-down list contains all output sections or memory segments within one output section. The “live” memory is where the overlay is stored before it is swapped into memory. The Overlay linking algorithm box permits one overlay algorithm— ALL_FIT. Expert Linker does not currently allow changes to this setting.

VisualDSP++ 5.0 Linker and Utilities Manual 4-65 Managing Object Properties

When ALL_FIT is used, the linker tries to fit all of the mapped objects into one overlay. The Placement box provides the following information: • Live Address—The starting address of the overlay • Run Address—The starting address where the overlay is swapped into memory at runtime • Size—The overlay’s size Click the Packing tab to specify byte packing order. The Browse button is only available after the overlay build and when the symbols are available. Clicking Browse opens the Browse Symbols dialog box. You can choose the address for the symbol group or let the linker choose the address.

4-66 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Managing Stack and Heap in Processor Memory

Expert Linker shows how much space is allocated for your program’s heap and stack. Figure 4-54 shows stack and heap output sections in the Memory Map pane. Right-click on either of them to display its properties.

Figure 4-54. Memory Map Window With Stack and Heap Sections

Use the Global Properties dialog box to select Show stack/heap usage (Figure 4-55). This option graphically displays the stack/heap usage in memory (Figure 4-56).

Figure 4-55. Global Properties – Selecting Stack and Heap Usage

VisualDSP++ 5.0 Linker and Utilities Manual 4-67 Managing Object Properties

The Expert Linker can: • Locate stacks and heaps and fill them with a marker value. This occurs after loading the program into a processor target. The stacks and heaps are located by their output section names, which may vary across processor families. • Search the heap and stack for the highest memory locations written to by the DSP program. This action occurs when the target halts after running the program. (assume the unused portion of the stack or heap starts here). The Expert Linker updates the memory map to show how much of the stack and heap are unused. Use this information to adjust the size of your stack and heap. This information helps make better use of the processor memory, so the stack and heap segments do not use too much memory.

4-68 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

Use the graphical view (View Mode -> Graphical Memory Map) to display stack and heap memory map blocks. Figure 4-56 shows a possible memory map after running a project program.

Figure 4-56. Graphical Memory Map Showing Stack and Heap Usage

VisualDSP++ 5.0 Linker and Utilities Manual 4-69 Managing Object Properties

Managing Shared Memory Properties

Specify the path and name of the file used by shared memory. This procedure assumes the Expert Linker window is open. To specify shared memory properties: 1. In the Memory Map pane, click the Shared Memory tab (located at the bottom of dialog box).

Figure 4-57. Shared Memory Tab

2. Right-click anywhere on the Memory Map pane. Note: Do not right-click on a memory segment, output section, input section, or overlay. 3. Choose Properties. The Shared Memory page of the Shared Memory Properties dialog box appears.

4-70 VisualDSP++ 5.0 Linker and Utilities Manual Expert Linker

4. In Output file name, specify the name of the output file for the shared memory. 5. In Processors sharing this memory, select the processors that share the file whose name appears in Output file name. Selecting a processor links its executable file against this shared memory file. 6. Optionally, click the Elimination tab (see “Managing Elimination Properties” on page 4-51) and specify options. 7. Click OK.

VisualDSP++ 5.0 Linker and Utilities Manual 4-71 Managing Object Properties

4-72 VisualDSP++ 5.0 Linker and Utilities Manual 5 MEMORY OVERLAYS AND ADVANCED LDF COMMANDS

This chapter describes memory management with the overlay functions as well as several advanced LDF commands used for multiprocessor-based systems. This chapter includes: • “Overview” on page 5-2 Provides an overview of Analog Devices processor’s overlay strategy • “Memory Management Using Overlays” on page 5-4 Describes memory management using the overlay functions • “Advanced LDF Commands” on page 5-29 Describes LDF commands that support memory management with overlay functions • “Linking Multiprocessor Systems” on page 5-39 Describes LDF commands that support the implementation of physical shared memory and building executable images for multiprocessor systems This chapter generally uses code examples for Blackfin processors. L If used, other processor’s code examples are marked accordingly.

VisualDSP++ 5.0 Linker and Utilities Manual 5-1 Overview

Overview

Analog Devices processors generally have a hierarchy of memory. The fast- est memory is the “internal” memory that is integrated with the processor on the same chip. For some processors, like Blackfin processors, there are two levels of internal memory (L1 and L2), with L1 memory being faster than L2 memory. Users can configure their system to include “external” memory, usually SDRAM or ROM that is connected to the part. Ideally, a program can fit in internal memory for optimal performance. Large programs need to be expanded to use external memory. When that happens, accessing code and data in slower memory can affect program performance. One way to address performance issues is to partition the program so that time-critical memory accesses are done using internal memory while parts of the program that are not time-critical can be placed in external memory. The placement of [program] sections into specific memory sections can be done using MEMORY{} and SECTION{} commands in the .ldf file. Another way to address performance issues is via memory architecture. Some memory architectures, for example, Blackfin architecture, have instruction and data cache. The processor can be configured to bring instructions and data into faster memory for fast processing. The third way to optimize performance is to use overlays. In an overlay system, code and data in slower memory is moved into faster memory when it is to be used. For architectures without cache, this method is the only way to run large parts of the program from fast internal memory. Even on processors with cache support, you may want to use overlays to have direct control of what is placed in internal memory for more deterministic behavior.

5-2 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

The overlay manager is a user-defined function responsible for ensuring that a required symbol (function or data) within an overlay is in run-time memory when it is needed. The transfer usually occurs using the direct memory access (DMA) capability of the processor. The overlay manager may also handle other advanced functionality described in “Introduction to Memory Overlays” on page 5-5 and “Overlay Managers” on page 5-7.

VisualDSP++ 5.0 Linker and Utilities Manual 5-3 Memory Management Using Overlays

Memory Management Using Overlays

To reduce DSP system costs, many applications employ processors with small amounts of on-chip memory and place much of the program code and data off-chip. The linker supports the linking of executable files for systems with overlay memory. Several applications notes (EE-Notes) on the Analog Devices Web site describe this technique in detail. This section describes the use of memory overlays. The topics are: • “Introduction to Memory Overlays” on page 5-5 • “Overlay Managers” on page 5-7 • “Memory Overlay Support” on page 5-8 • “Example – Managing Two Overlays” on page 5-13 • “Linker-Generated Constants” on page 5-15 • “Overlay Word Sizes” on page 5-16 • “Storing Overlay ID” on page 5-20 • “Overlay Manager Function Summary” on page 5-20 • “Reducing Overlay Manager Overhead” on page 5-21 • “Using PLIT{} and Overlay Manager” on page 5-25 The following LDF commands facilitate overlay features. • “OVERLAY_GROUP{}” on page 5-29 • “PLIT{}” on page 5-34

5-4 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

Introduction to Memory Overlays

Memory overlays support applications that cannot fit the program instructions into the processor’s internal memory. In such cases, program instructions are partitioned and stored in external memory until they are required for program execution. These partitions are memory overlays, and the routines that call and execute them are called overlay managers. Overlays are “many to one” memory-mapping systems. Several overlays may “live” (be stored) in unique locations in external memory, but “run” (execute) in a common location in internal memory. Throughout the following description, the overlay storage location is referred to as the “live” location, and the internal location where instructions are executed is referred to as the “run” (run-time) space.

Overlay functions are written to overlay files (.ovl), which are specified as one type of linker executable output file. The loader can read .ovl files to generate an .ldr file. Figure 5-1 demonstrates the concept of memory overlays. The two memory spaces are: internal and external. The external memory is partitioned into the live space for four overlays. The internal memory contains the

VisualDSP++ 5.0 Linker and Utilities Manual 5-5 Memory Management Using Overlays

main program, an overlay manager function, and two memory segments reserved for execution of overlay program instructions (run space).

External Memory Internal Memory

Overlay 1 FUNC_A Main: call FUNC_H call .plt_FUNC_A

Overlay 2 FUNC_B FUNC_C Overlay Manager

Overlay 3 FUNC_D FUNC_E Overlay 1 and 2 Runtime Memory

Overlay 4 FUNC_F FUNC_G Overlay 3 and 4 Runtime Memory

Figure 5-1. Memory Overlays

In this example, overlays 1 and 2 share the same run-time location within internal memory, and overlays 3 and 4 also share a common run-time memory. When FUNC_B is required, the overlay manager loads overlay 2 to the location in internal memory where overlay 2 is designated to run. When FUNC_D is required, the overlay manager loads overlay 3 into its designated run-time memory. The transfer is typically implemented with the processor’s direct memory access (DMA) capability. The overlay manager can also handle advanced functionality, such as checking whether the requested overlay is already in run-time memory, executing another function while loading an overlay, and tracking recursive overlay function calls.

5-6 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

Overlay Managers

An overlay manager is a user-definable routine responsible for loading a referenced overlay function or data buffer into internal memory (run space). This task is accomplished with linker-generated constants and PLIT{} commands. Linker-generated constants inform the overlay manager of the overlay’s live address, where the overlay resides for execution. The number of words in the overlay PLIT{} commands inform the overlay manager of the requested overlay and the run-time address of the referenced symbol. An overlay manager’s main objective is to transfer overlays to a run-time location when required. Overlay managers may also: • Set up a stack to store register values • Check whether a referenced symbol has already been transferred into its run-time space as a result of a previous reference

If the overlay is already in internal memory, the overlay transfer is bypassed and execution of the overlay routine begins immediately. • Load an overlay while executing a function from a second overlay (or a non-overlay function) You may require an overlay manager to perform other specialized tasks to satisfy the special needs of a given application. Overlay managers are application-specific and must be developed by the user.

Breakpoints on Overlays

The debugger relies on the presence of the _ov_start and _ov_end symbols to support breakpoints on overlays. These symbols should appear in the user’s overlay manager for debugger support of overlays. The symbol manager sets a silent breakpoint at each symbol.

VisualDSP++ 5.0 Linker and Utilities Manual 5-7 Memory Management Using Overlays

The more important of the two symbols is the breakpoint at _ov_end. Code execution in the overlay manager passes through this location once an overlay is fully swapped in. At this point, the debugger may probe the target to determine which overlays are in context. The symbol manager now sets any breakpoints requested on the overlays and resumes execution.

The second breakpoint is at symbol _ov_start. The label _ov_start is defined in the overlay manager (in code always executed immediately before the transfer of a new overlay begins). The breakpoint disables all of the overlays in the debugger—the idea being that while the target is running in the overlay manager, the target is “unstable” in the sense that the debugger should not rely on the overlay information it may gather since the target is “in flux”. The debugger still functions without this breakpoint, but there may be inconsistencies while overlays are being moved in and out.

Memory Overlay Support

The overlay support provided by the DSP tools includes: • Specification of the live and run locations of each overlay • Generation of constants • Redirection of overlay function calls to a jump table

Overlay support is partially user-designed in the .ldf file. You specify which overlays share run-time memory and which memory segments establish the “live” and “run” space.

Listing 5-1 shows the portion of an .ldf file that defines two overlays. This overlay declaration configures the two overlays to share a common run-time memory space. The syntax for the OVERLAY_INPUT{} command is described in “OVERLAY_INPUT{overlay_commands}” on page 3-70.

5-8 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

In this code example, OVLY_one contains FUNC_A and lives in memory segment ovl_live; OVLY_two contains functions FUNC_B and FUNC_C and also lives in memory segment ovl_live.

Listing 5-1. Overlay Declaration in an LDF

.dxe_code { OVERLAY_INPUT { OVERLAY_OUTPUT (OVLY_one.ovl) INPUT_SECTIONS (FUNC_A.doj(program)) } >ovl_live

OVERLAY_INPUT { OVERLAY_OUTPUT (OVLY_two.ovl) INPUT_SECTIONS (FUNC_B.doj(program) FUNC_C.doj(sec_code)) } >ovl_live } >ovl_run

The common run-time location shared by overlays OVLY_one and OVLY_two is within the ovl_run memory segment.

The .ldf file configures the overlays and provides the information necessary for the overlay manager to load the overlays. The information includes the following linker-generated overlay constants (where # is the overlay ID).

_ov_startaddress_# _ov_endaddress_# _ov_size_# _ov_word_size_run_# _ov_word_size_live_# _ov_runtimestartaddress_#

Each overlay has a word size and an address, which is used by the overlay manager to determine where the overlay resides and where it is executed.

VisualDSP++ 5.0 Linker and Utilities Manual 5-9 Memory Management Using Overlays

_ov_word_size_run_# and _ov_word_size_live_# are both in terms of words, _ov_size_# specifies the total size in bytes. Overlay “live” and “run” word sizes differ when internal memory and external memory widths differ. A system containing either 16-bit-wide or 32-bit-wide external memory requires data packing to store an overlay containing instructions. The Blackfin processor architecture supports byte addressing that L uses 16-, 32-, or 64-bit opcodes. Thus, no data packing is required. Redirection In addition to providing constants, the linker replaces overlay symbol references to the overlay manager within your code. Redirection is accomplished by means of a procedure linkage table (PLIT), which is essentially a jump table that executes user-defined code and then jumps to the overlay manager. The linker replaces an overlay symbol reference (function call) with a jump to a location in the PLIT.

You must define PLIT code within the .ldf file. This code prepares the overlay manager to handle the overlay that contains the referenced symbol. The code initializes registers to contain the overlay ID and the referenced symbol’s run-time address. The linker reserves one word (or two bytes in Blackfin processors) L at the top of an overlay to house the overlay ID. The following is an example call instruction to an overlay function: CALL FUNC_A;; /* Call to function in overlay */

If FUNC_A is in an overlay, the linker replaces the function call with the following instruction: CALL .plt_FUNC_A; / * Call to PLIT entry */

.plt_FUNC_A is the entry in the PLIT that contains defined instructions. These instructions prepare the overlay manager to load the overlay containing FUNC_A. The instructions executed in the PLIT are specified within

5-10 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

the .ldf file. The user must supply the PLIT code to match the overlay manager.

Listing 5-2 is an example PLIT definition from an .ldf file, where register R0 is set to the value of the overlay ID that contains the referenced symbol and register R1 is set to the run-time address of the referenced symbol. The last instruction branches to the overlay manager that uses the initialized registers to determine which overlay to load (and where to jump to execute the called overlay function).

Listing 5-2. PLIT Definitions in LDF

PLIT // Blackfin PLIT { R0.l = PLIT_SYMBOL_OVERLAYID; R1.h = PLIT_SYMBOL_ADDRESS; R1.l = PLIT_SYMBOL_ADDRESS; JUMP OverlayManager; }

The linker expands the PLIT definition into individual entries in a table. An entry is created for each overlay symbol as shown in Listing 5-2. The redirection function calls the PLIT table for overlays 1 and 2 (Figure 5-2).

VisualDSP++ 5.0 Linker and Utilities Manual 5-11 Memory Management Using Overlays

For each entry, the linker replaces the generic assembly instructions with specific instructions (where applicable).

Overlay 1 Overlay 2

FUNC_A FUNC_B FUNC_C

Internal Memory

Main: Plit_table call .plt_FUNC_A .plt_FUNC_A R0.L = 0x00001; . R1.H = 0x00000; . R1.L = 0x22000; . jumpOverlayManager;

call .plt_FUNC_C .plt_FUNC_B R0.L = 0x00002; call .plt_FUNC_B R1.H = 0x00000; call .plt_FUNC_C. R1.L = 0x22000; . jumpOverlayManager; . .plt_FUNC_C R0.L = 0x00002; R1.H = 0x00000; R1.L = 0x23000; jumpOverlayManager;

Figure 5-2. Expanded PLIT Table (for TigerSHARC Processors)

For example, the first PLIT entry in Figure 5-2 is for the overlay symbol FUNC_A. The linker replaces the constant name PLIT_SYMBOL_OVERLAYID with the ID of the overlay containing FUNC_A. The linker also replaces the constant name PLIT_SYMBOL_ADDRESS with the run-time address of FUNC_A. When the overlay manager is called via the jump instruction of the PLIT table, R0 contains the referenced function’s overlay ID and R1 contains the referenced function’s run-time address. The overlay manager uses the overlay ID and run-time address to load and execute the referenced function.

5-12 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

Example – Managing Two Overlays

Overlay manager are user-written, and the following is an example of what an overlay manager can do. This example has two overlays, each containing two functions. Overlay 1 contains the functions fft_first_two_stages and fft_last_stage. Overlay 2 contains functions fft_middle_stages and fft_next_to_last. For examples of overlay manager source code, refer to the example programs shipped with the development software. The overlay manager: • Creates and maintains a stack for the registers it uses • Determines whether the referenced function is in internal memory • Sets up a DMA transfer • Executes the referenced function

Several code segments for the .ldf file and the overlay manager follow with appropriate explanations.

Listing 5-3. FFT Overlay Example 1

{ OVERLAY_INPUT { OVERLAY_OUTPUT (fft_one.ovl) INPUT_SECTIONS ( Fft_1st_last.doj(program) ) } > ovl_livee // Overlay to live in section ovl_live

OVERLAY_INPUT { OVERLAY_OUTPUT (fft_two.ovl) INPUT_SECTIONS ( Fft_mid.doj(program) )

VisualDSP++ 5.0 Linker and Utilities Manual 5-13 Memory Management Using Overlays

} > ovl_live // Overlay to live in section ovl_live } > ovl_run

The two defined overlays (fft_one.ovl and fft_two.ovl) live in memory segment ovl_live (defined by the MEMORY{} command), and run in section ovl_run. All instruction and data defined in the program memory segment within the Fft_1st_last.doj file are part of the fft_one.ovl overlay. All instructions and data defined in program within the file Fft_mid.doj are part of overlay fft_two.ovl. The result is two functions within each overlay.

The first and the last called functions are in overlay fft_one. The two middle functions are in overlay fft_two. When the first function (fft_one) is referenced during code execution, overlay id=1 is transferred to internal memory. When the second function (fft_two) is referenced, overlay id=2 is transferred to internal memory. When the third function (in overlay fft_two) is referenced, the overlay manager recognizes that it is already in internal memory and an overlay transfer does not occur. To verify whether an overlay is in internal memory, place the overlay ID of this overlay into a register (for example, P0) and compare this value to the overlay ID of each loaded overlay. This is done by loading these overlay values into a register (for example, R1).

/* Is overlay already in internal memory? */ CC = p0 == p1; /* If so, do not transfer it in. */ if CC jump skipped_DMA_setup;

Finally, when the last function (fft_one) is referenced, overlay id=1 is again transferred to internal memory for execution. The following code segment calls the four FFT functions.

fftrad2: call fft_first_2_stages;; call fft_middle_stages;;

5-14 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

call fft_next_to_last;; call fft_last_stage;; wait: NOP;; jump wait;;

The linker replaces each overlay function call with a call to the appropriate entry in the PLIT. For this example, only three instructions are placed in each entry of the PLIT.

PLIT { R0.l = PLIT_SYMBOL_OVERLAYID; R1.h = PLIT_SYMBOL_ADDRESS; R1.l = PLIT_SYMBOL_ADDRESS; JUMP OverlayManager; }

Register R0 contains the overlay ID with the referenced symbol, and register R1 contains the run-time address of the referenced symbol. The final instruction jumps to the starting address of the overlay manager. The overlay manager uses the overlay ID in conjunction with the overlay constants generated by the linker to transfer the proper overlay into internal memory. Once the transfer is complete, the overlay manager jumps to the address of the referenced symbol stored in R1.

Linker-Generated Constants

The following constants, which are generated by the linker, are used by the overlay manager.

.EXTERN _ov_startaddress_1; .EXTERN _ov_startaddress_2; .EXTERN _ov_endaddress_1; .EXTERN _ov_endaddress_2; .EXTERN _ov_size_1;

VisualDSP++ 5.0 Linker and Utilities Manual 5-15 Memory Management Using Overlays

.EXTERN _ov_size_2; .EXTERN _ov_word_size_run_1; .EXTERN _ov_word_size_run_2; .EXTERN _ov_word_size_live_1; .EXTERN _ov_word_size_live_2; .EXTERN _ov_runtimestartaddress_1; .EXTERN _ov_runtimestartaddress_2;

The constants provide the following information to the overlay manager. • Overlay sizes (both run-time word sizes and live word sizes) • Starting address of the “live” space • Starting address of the “run” space

Overlay Word Sizes

Each overlay has a word size and an address, which the overlay manager uses to determine where the overlay resides and where it is executed. Table 5-1 shows the linker-generated constants and examples of processor-specific addresses.

Table 5-1. Linker-Generated Constants and Processor Addresses

Constant Blackfin Processors

_ov_startaddress_1 0x00000000

_ov_startaddress_2 0x00000010

_ov_endaddress_1 0x0000000F

_ov_endaddress_2 0x0000001F

_ov_word_size_run_1 0x00000010

_ov_word_size_run_2 0x00000010

_ov_word_size_live_1 0x00000010

_ov_word_size_live_2 0x00000010

5-16 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

Table 5-1. Linker-Generated Constants and Processor Addresses (Cont’d)

Constant Blackfin Processors

_ov_runtimestartaddress_1 0xF0001000

_ov_runtimestartaddress_2 0xF0001000

The overlay manager places the constants in arrays as shown in Figure 5-3 and Figure 5-4. The arrays are referenced by using the overlay ID as the index to the array. The index or ID is stored in a Modify register (Jn/Kn for TigerSHARC processors and M# for SHARC and Blackfin processors), and the beginning address of the array is stored in the Index register (Jm/Km for TigerSHARC processors and I# for SHARC and Blackfin processors).

.VAR liveAddresses[2] = _ov_startaddress_1, _ov_startaddress_2; .VAR runAddresses[2] = _ov_runtimestartaddress_1, _ov_runtimestartaddress_2; .VAR runWordSize[2] = _ov_word_size_run_1, _ov_word_size_run_2;

VisualDSP++ 5.0 Linker and Utilities Manual 5-17 Memory Management Using Overlays

.VAR liveWordSize[2] = _ov_word_size_live_1, _ov_word_size_live_2;

External Memory Internal Memory

Address 0x04000000 Address 0x00001000 Overlay 1 (24 x 32-bits) Overlay Runtime Memory (40 x 48-bits) FUNC_A

0x0400007F Overlay 1 Overlay 2 16 x 48 bits 40 x 48 bits 0x04000080 Overlay 2 (60 x 32-bits)

FUNC_B FUNC_C

0x040000FF

Figure 5-3. TigerSHARC Overlay Live and Run Memory Sizes

5-18 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

External Memory Internal Memory

Address 0x2 0000 Overlay 1 Address 0x8800 Overlay Runtime Memory (24 x 32-bits) (40 x 48-bits)

FUNC_A 0x2 0017

Overlay 1 Overlay 2 0x2 0018 16 x 48 bits 40 x 48 bits

Overlay 2

(60 x 32-bits)

FUNC_B

FUNC_C 0x2 0053

Figure 5-4. SHARC Overlay Live and Run Memory Sizes

Figure 5-4 shows the difference between overlay “live” and “run” size in SHARC processor memory: • Overlays 1 and 2 are instruction overlays with a run word width of 48 bits. • Because external memory is 32 bits, the live word size is 32 bits. • Overlay 1 contains one function with 16 instructions. Overlay 2 contains two functions with a total of 40 instructions. • The “live” word size for overlays 1 and 2 are 24 and 60 words, respectively. • The “run” word size for overlay 1 and 2 are 16 and 40 words, respectively.

VisualDSP++ 5.0 Linker and Utilities Manual 5-19 Memory Management Using Overlays

Storing Overlay ID

The overlay manager stores the ID of an overlay currently residing in internal memory. When an overlay is transferred to internal memory, the overlay manager stores the overlay ID in internal memory in the buffer labeled ov_id_loaded. Before another overlay is transferred, the overlay manager compares the required overlay ID with the ID stored in the ov_id_loaded buffer. If they are equal, the required overlay is already in internal memory and a transfer is not required. The PC is sent to the proper location to execute the referenced function. If they are not equal, the value in ov_id_loaded is updated and the overlay is transferred into its internal run space via DMA. On completion of the transfer, the overlay manager restores register values from the run-time stack, flushes the cache, and then jumps the PC to the run-time location of the referenced function. It is very important to flush the cache before moving the PC to the referenced function. Otherwise, when code is replaced or modified, incorrect code execution may occur. If the program sequencer searches the cache for an instruction and an instruction from the previous overlay is in the cache, that instruction may be executed because the expected cache miss is not received.

Overlay Manager Function Summary

In summary, the overlay manager routine: • Maintains a run-time stack for registers being used by the overlay manager • Compares the requested overlay’s ID with that of the previously loaded overlay (stored in the ov_id_loaded buffer) • Sets up the DMA transfer of the overlay (if it is not already in internal memory) • Jumps the PC to the run-time location of the referenced function

5-20 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

These are the basic tasks that are performed by an overlay manager. More sophisticated overlay managers may be required for individual applications.

Reducing Overlay Manager Overhead

The example in this section incorporates the ability to transfer one overlay to internal memory while the core executes a function from another overlay. Instead of the core sitting idle while the overlay DMA transfer occurs, the core enables the DMA, and then begins executing another function. This example uses the concept of overlay function loading and executing. A function load is a request to load the overlay function into internal memory but not execute the function. A function execution is a request to execute an overlay function that may or may not be in internal memory at the time of the execution request. If the function is not in internal memory, a transfer must occur before execution. In several circumstances, an overlay transfer can be in progress while the core is executing another task. Each circumstance can be labeled as deterministic or non-deterministic. A deterministic circumstance is one where you know exactly when an overlay function is required for execution. A non-deterministic circumstance is one where you cannot predict when an overlay function is required for execution. For example, a deterministic application may consist of linear flow code except for function calls. A non-deterministic example is an application with calls to overlay functions within an interrupt service routine (ISR) where the interrupt occurs randomly. The example provided by the software contains deterministic overlay function calls. The time of overlay function execution requests are known as the number of cycles required to transfer an overlay. Therefore, an overlay function load request can be placed to complete the transfer by the time the execution request is made. The next overlay transfer (from a load

VisualDSP++ 5.0 Linker and Utilities Manual 5-21 Memory Management Using Overlays

request) can be enabled by the core, and the core can execute the instructions leading up to the function execution request. Since the linker handles all overlay symbol references in the same way (jump to PLIT table and then overlay manager), the overlay manager must distinguish between a symbol reference requesting the load of an overlay function and a symbol reference requesting the execution of an overlay function. In the example, the overlay manager uses a buffer in memory as a flag to indicate whether the function call (symbol reference) is a load or an execute request. The overlay manager first determines whether the referenced symbol is in internal memory. If not, it sets up the DMA transfer. If the symbol is not in internal memory and the flag is set for execution, the core waits for the transfer to complete (if necessary) and then executes the overlay function. If the symbol is set for load, the core returns to the instructions immediately following the location of the function load reference. Every overlay function call requires initializing the load/execute flag buffer. Here, the function calls are delayed branch calls. The two slots in the delayed branch contain instructions to initialize the flag buffer. Register j4 is set to the value placed in the flag buffer, and the value in j4 is stored in memory; 1 indicates a load, and 0 indicates an execution call. At each overlay function call, the load buffer must be updated. The following code is from the main FFT subroutine. Each of the four function calls are execution calls so the pre-fetch (load) buffer is set to zero. The flag buffer in memory is read by the overlay manager to determine whether the function call is a load or an execution call.

R0 = 0 (Z); p0.h = prefetch; p0.l = prefetch; [P0] = R0; call fft_first_2_stages; R0 = 0 (Z);

5-22 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

p0.h = prefetch; p0.l = prefetch; [P0] = R0; call fft_middle_stages; R0 = 0 (Z); p0.h = prefetch; p0.l = prefetch; [P0] = R0; call fft_next_to_last; R0 = 0 (Z); p0.h = prefetch; p0.l = prefetch; [P0] = R0; call fft_last_stage;

The next set of instructions represents a load function call.

R0 = 1 (Z); p0.h = prefetch; p0.l = prefetch; [P0] = R0; /* Set prefetch flag to 1 to indicate a load */ call fft_middle_stages; /* Pre-loads the function into the */ /* overlay run memory. */

The code executes the first function and transfers the second function and so on. In this implementation, each function resides in a unique overlay and requires two run-time locations. While one overlay loads into one run-time location, a second overlay function executes in another run-time location.

VisualDSP++ 5.0 Linker and Utilities Manual 5-23 Memory Management Using Overlays

The following code segment allocates the functions to overlays and forces two run-time locations.

OVERLAY_GROUP1 { OVERLAY_INPUT { ALGORITHM(ALL_FIT) OVERLAY_OUTPUT(fft_one.ovl) INPUT_SECTIONS( Fft_ovl.doj (program) ) } >ovl_code // Overlay to live in section ovl_code OVERLAY_INPUT { ALGORITHM(ALL_FIT) OVERLAY_OUTPUT(fft_three.ovl) INPUT_SECTIONS( Fft_ovl.doj (program) ) } >ovl_code // Overlay to live in section ovl_code } > mem_code

OVERLAY_MGR { INPUT_SECTIONS(ovly_mgr.doj(program)) } > mem_code

OVERLAY_GROUP2 { OVERLAY_INPUT { ALGORITHM(ALL_FIT) OVERLAY_OUTPUT(fft_two.ovl) INPUT_SECTIONS( Fft_ovl.doj(program) ) } >ovl_code // Overlay to live in section ovl_code OVERLAY_INPUT { ALGORITHM(ALL_FIT) OVERLAY_OUTPUT(fft_last.ovl) INPUT_SECTIONS( Fft_ovl.doj(program) )

5-24 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

} >ovl_code // Overlay to live in section ovl_code } > mem_code

The first and third overlays share one run-time location, and the second and fourth (last) overlays share the second run-time location. Additional instructions are included to determine whether the function call is a load or an execution call. If the function call is a load, the overlay manager initiates the DMA transfer and then jumps the PC back to the location where the call was made. If the call is an execution call, the overlay manager determines whether the overlay is currently in internal memory. If so, the PC jumps to the run-time location of the called function. If the overlay is not in internal memory, a DMA transfer is initiated and the core waits for the transfer to complete. The overlay manager pushes the appropriate registers on the run-time stack. It checks whether the requested overlay is currently in internal memory. If not, the overlay manager sets up the DMA transfer. It then checks whether the function call is a load or an execution call. If it is a load call, the overlay manager begins the transfer and returns the PC back to the instruction following the call. If it is an execution call, the core is idle until the transfer is completed (if the transfer was necessary). The PC then jumps to the run-time location of the function. Specific applications may require specific code modifications, L which may eliminate some instructions. For instance, if your application allows the free use of registers, you may not need a run-time stack.

Using PLIT{} and Overlay Manager

The PLIT{} command inserts assembly instructions that handle calls to functions in overlays. The instructions are specific to an overlay and are executed each time a call to a function in that overlay is detected.

VisualDSP++ 5.0 Linker and Utilities Manual 5-25 Memory Management Using Overlays

Refer to “PLIT{}” on page 5-34 for basic syntax information. Refer to “Introduction to Memory Overlays” on page 5-5 for detailed information on overlays. Figure 5-5 shows the interaction between a PLIT and an overlay manager.

Non-Overlay Memory main() { int (*pf)() = X; Y(); }

/* PLIT & overlay manager handle calls, using the PLIT to resolve calls and load overlays as needed */

.plt_X: call OM

.plt_Y: call OM

Overlay 1 Storage X() {...} // function X defined

Overlay 2 Storage Y() {...} // function Y defined

Run-time Overlay Memory // currently loaded overlay

Figure 5-5. PLITs and Overlay Memory; main() Calls to Overlays

To make this kind of interaction possible, the linker generates special symbols for overlays. These overlay symbols are:

• _ov_startaddress_#

• _ov_endaddress_#

• _ov_size_#

• _ov_word_size_run_#

5-26 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

• _ov_word_size_live_#

• _ov_runtimestartaddress_#

The # indicates the overlay number. Overlay numbers start at 1 (not 0) to avoid confusion when these L elements are placed into an array or buffer used by an overlay manager. The two functions in Figure 5-5 describe different overlays. By default, the linker generates PLIT code only when an unresolved function reference is resolved to a function definition in overlay memory.

The main function calls functions X() and Y(), which are defined in overlay memory. Because the linker cannot resolve these functions locally, the linker replaces the symbols X and Y with .plit_X and .plit_Y. Unresolved references to X and Y are resolved to .plit_X and .plit_Y. When the reference and the definition reside in the same executable file, the linker does not generate PLIT code. However, you can force the linker to output a PLIT, even when all references can be resolved locally. The PLIT code sets up data for the overlay manager, which first loads the overlay that defines the desired symbol, and then branches to that symbol.

Inter-Overlay Calls PLITs resolve inter-processor overlay calls, as shown in Figure 5-6, for systems that permit one processor to access the memory of another processor. When one processor calls into another processor’s overlay, the call increases the size of the .plit section in the executable file that manages the overlay. The linker resolves all references to variables in overlays, and the PLIT lets an overlay manager handle the overhead of loading and unloading overlays.

VisualDSP++ 5.0 Linker and Utilities Manual 5-27 Memory Management Using Overlays

Placing global variables in non-overlay memory optimizes overlays. L This action ensures that the proper overlay is loaded before a global variable is referenced.

Inter-Processor Calls PLITs resolve inter-processor overlay calls, as shown in Figure 5-6, for systems that permit one processor to access the memory of another processor.

Processor P1 Processor P2 Non-overlay Memory Non-overlay Memory

main() { P2_Overlay_Manager() .plt_foo(); { } // manager routines }

/* PLIT & overlay manager handle calls using the PLIT to resolve calls and load overlays as needed */ .plt_foo: call P2_Overlay_Manager

Processor P2 Overlay Storage

P2 Overlay foo() { ... }

Processor P2 Overlay Memory

// current overlay

Figure 5-6. PLITs and Overlay Memory – Inter-Processor Calls

5-28 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

When one processor calls into another processor’s overlay, the call increases the size of the .plit section in the executable file that manages the overlay. The linker resolves all references to variables in overlays, and the PLIT lets an overlay manager handle the overhead of loading and unloading overlays. Not putting global variables in overlays optimizes overlays. This L action ensures that the proper overlay is loaded before a global is referenced.

Advanced LDF Commands

Commands in the .ldf file define the target system and specify the order in which the linker processes output for that system. The LDF commands operate within a scope, which influences the operation of other commands that appear within the range of that scope. The following LDF commands support advanced memory management functions, overlays, and shared memory features. • “OVERLAY_GROUP{}” on page 5-29 • “PLIT{}” on page 5-34 For detailed information on multiprocessor-related LDF commands, refer to “Linking Multiprocessor Systems” on page 5-39.

OVERLAY_GROUP{}

The OVERLAY_GROUP{} command provides legacy support. This command is deprecated and is not recommended for use. When running the linker, the following warning may occur.

VisualDSP++ 5.0 Linker and Utilities Manual 5-29 Advanced LDF Commands

[Warning li2534] More than one overlay group or explicit OVERLAY_GROUP command is detected in the output section 'seg_data1'. Create a separate output section for each group of overlays. Expert Linker makes the change automatically upon reading the LDF.

Memory overlays support applications whose program instructions and data do not fit in the internal memory of the processor.

Overlays may be grouped or ungrouped. Use the OVERLAY_INPUT{} command to support ungrouped overlays. Refer to “Memory Overlay Support” on page 5-8 for a detailed description of overlay functionality.

Overlay declarations syntactically resemble the SECTIONS{} commands. They are portions of SECTIONS{} commands.

The OVERLAY_GROUP{} command syntax is:

OVERLAY_GROUP { OVERLAY_INPUT { ALGORITHM(ALL_FIT) OVERLAY_OUTPUT() INPUT_SECTIONS() } }

In the simplified examples in Listing 5-4 and Listing 5-5, the functions are written to overlay (.ovl) files. Whether functions are disk files or memory segments does not matter (except to the DMA transfer that brings them in). Overlays are active only while being executed in run-time memory, which is located in the program memory segment.

5-30 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

Ungrouped Overlay Execution In Listing 5-4, as the FFT progresses and overlay functions are called in turn, they are brought into run-time memory in sequence as four function transfers. Figure 5-7 shows the ungrouped overlays. “Live” locations reside in several different memory segments. The L linker outputs the executable overlay (.ovl) files while allocating destinations for them in the program section.

fft_one.ovl Main: call OVERLAY_INPUT Overlay call {fft_one.ovl} fft_two.ovl Overlay OVERLAY_INPUT Overlay Manager {fft_two.ovl} fft_three.ovl Overlay Run-time OVERLAY_INPUT Overlay Memory {fft_three.ovl} fft_last.ovl OVERLAY_INPUT Overlay {fft_last.ovl}

Figure 5-7. Example of Overlays – Not Grouped

Listing 5-4. LDF Overlays – Not Grouped

// This is part of the SECTIONS{} command for processor P0 // Declare which functions reside in which overlay. // The overlays have been split into different segments // in one file, or into different files. // The overlays declared in this section (seg_pmco) // will run in segment seg_pmco.

OVERLAY_INPUT { // Overlays to live in section ovl_code ALGORITHM ( ALL_FIT ) OVERLAY_OUTPUT ( fft_one.ovl)

VisualDSP++ 5.0 Linker and Utilities Manual 5-31 Advanced LDF Commands

INPUT_SECTIONS ( Fft_1st.doj(program) ) } >ovl_code

OVERLAY_INPUT { ALGORITHM ( ALL_FIT ) OVERLAY_OUTPUT ( fft_two.ovl) INPUT_SECTIONS ( Fft_2nd.doj(program) ) } >ovl_code

OVERLAY_INPUT { ALGORITHM ( ALL_FIT ) OVERLAY_OUTPUT ( fft_three.ovl) INPUT_SECTIONS ( Fft_3rd.doj(program) ) } >ovl_code

OVERLAY_INPUT { ALGORITHM ( ALL_FIT ) OVERLAY_OUTPUT ( fft_last.ovl) INPUT_SECTIONS ( Fft_last.doj(program) ) } >ovl_code

5-32 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

Grouped Overlay Execution Figure 5-8 demonstrates grouped overlays.

OVERLAY_GROUP{ fft_one.ovl OVERLAY_INPUT{ overlay Main: call fft_one.ovl} call OVERLAY_INPUT{ fft_two.ovl} fft_two.ovl } overlay OVERLAY_GROUP{ Overlay Manager OVERLAY_INPUT{ fft_three.ovl} fft_three.ovl OVERLAY_INPUT{ overlay fft_last.ovl} Overlay Group 1 } Runtime fft_last.ovl Memory overlay Overlay Group 2 Runtime Memory

Figure 5-8. Example of Overlays – Grouped

Listing 5-5 shows a different implementation of the same algorithm. The overlay functions are grouped in pairs. Since all four pairs of routines reside simultaneously, the processor executes both routines before paging.

Listing 5-5. LDF Overlays – Grouped

OVERLAY_GROUP { // Declare first overlay group OVERLAY_INPUT { // Overlays to live in section ovl_code ALGORITHM ( ALL_FIT ) OVERLAY_OUTPUT ( fft_one.ovl) INPUT_SECTIONS ( Fft_1st.doj(program) ) } >ovl_code OVERLAY_INPUT { ALGORITHM ( ALL_FIT )

VisualDSP++ 5.0 Linker and Utilities Manual 5-33 Advanced LDF Commands

OVERLAY_OUTPUT ( fft_two.ovl) INPUT_SECTIONS ( Fft_mid.doj(program) ) } >ovl_code } OVERLAY_GROUP { // Declare second overlay group OVERLAY_INPUT { // Overlays to live in section ovl_code ALGORITHM ( ALL_FIT ) OVERLAY_OUTPUT ( fft_three.ovl) INPUT_SECTIONS ( Fft_last.doj(program) ) } >ovl_code OVERLAY_INPUT { ALGORITHM ( ALL_FIT ) OVERLAY_OUTPUT ( fft_last.ovl) INPUT_SECTIONS ( Fft_last.doj(program) ) } >ovl_code }

PLIT{}

The linker resolves function calls and variable accesses (both direct and indirect) across overlays. This task requires the linker to generate extra code to transfer control to a user-defined routine (an overlay manager) that handles the loading of overlays. Linker-generated code goes in a special section of the executable file, which has the section name .PLIT.

The PLIT{} command in an .ldf file inserts assembly instructions that handle calls to functions in overlays. The assembly instructions are specific to an overlay and are executed each time a call to a function in that overlay is detected.

The PLIT{} command provides a template from which the linker generates assembly code when a symbol resolves to a function in overlay memory. The code typically handles a call to a function in overlay memory by calling an overlay memory manager. Refer to “Memory Overlay Support” on page 5-8 for a detailed description of overlay and PLIT functionality.

5-34 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

A PLIT{} command may appear in the global LDF scope, within a PROCESSOR{} command, or within a SECTIONS{} command. For an example of using a PLIT{} command, see “Using PLIT{} and Overlay Manager” on page 5-25.

When writing the PLIT{} command in the .ldf file, the linker generates an instance of the PLIT, with appropriate values for the parameters involved, for each symbol defined in overlay code.

PLIT Syntax

Figure 5-9 shows the general syntax of the PLIT{} command and indicates how the linker handles a symbol (symbol) local to an overlay function.

PLIT{plit_commands}

instruction symbol = PLIT_SYMBOL_OVERLAYID [symbol] symbol = PLIT_SYMBOL_ADDRESS symbol = PLIT_DATA_OVERLAY_ID

Figure 5-9. PLIT{} Command Syntax Tree

Parts of the PLIT{} command are:

• instruction – None, one, or multiple assembly instructions. The instructions may occur in any reasonable order in the command structure and may precede or follow symbols. The following

VisualDSP++ 5.0 Linker and Utilities Manual 5-35 Advanced LDF Commands

two constants contain information about symbol and the overlay in which it occurs. You must supply instructions to handle that information.

• PLIT_SYMBOL_OVERLAYID – Returns the overlay ID

• PLIT_SYMBOL_ADDRESS – Returns the absolute address of the resolved symbol in run-time memory

Command Evaluation and Setup The linker first evaluates the sequence of assembly code in each plit_command. Each line is passed to a processor-specific assembler, which supplies values for the symbols and expressions. After evaluation, the linker places the returned bytes into the .plit output section and manages the addressing in that output section. To help write an overlay manager, the linker generates PLIT constants for each symbol in an overlay. Data can be overlaid, just like code. If an overlay-resident function calls for additional data overlays, include an instruction for finding them. After the setup and variable identification are completed, the overlay itself is brought (via DMA transfer) into run-time memory. This process is controlled by assembly code called an overlay manager.

The branch instruction, such as JUMP OverlayManager, is normally L the last instruction in the PLIT{} command.

Overlay PLIT Requirements and PLIT Examples

Both the .plit output section (allocating space for PLIT) and the PLIT{} command are necessary when specifying PLIT for overlays. The .ldf file must allocate space in memory to hold PLITs built by the linker. Typi- cally, that memory resides in the program code memory segment.

5-36 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

No input section is associated with the .plit output section. The .ldf file allocates space for linker-generated routines, which do not contain (input) data objects. A typical LDF declaration for that purpose is:

// ... [In the SECTIONS command for Processor P0] // Plit code is to reside and run in mem_program segment .plit {} > mem_program

This segment allocation does not take any parameters. You write the structure of this command according to the PLIT syntax. The linker creates an instance of the command for each symbol that resolves to an overlay. The linker stores each instance in the .plit output section, which becomes part of the program code’s memory segment.

A PLIT{} command may appear in the global LDF scope, within a PROCESSOR{} command, or within a SECTIONS{} command. Simple PLIT – States are not Saved A simple PLIT merely copies the symbol’s address and overlay ID into registers and jumps to the overlay manager. The following fragment is extracted from the global scope (just after the MEMORY{} command) of sample fft_group.ldf. Verify that the contents of P0 and P1 are either safe or irrelevant. For example,

PLIT { P0 = PLIT_SYMBOL_OVERLAY_ID; P1.L = PLIT_SYMBOL_ADDRESS; P1.H = PLIT_SYMBOL_ADDRESS; JUMP _OverlayManager; }

VisualDSP++ 5.0 Linker and Utilities Manual 5-37 Advanced LDF Commands

As a general rule, minimize overlay transfer traffic. Improve performance by designing code to ensure overlay functions are imported and use minimal (or no) reloading.

PLIT – Summary A PLIT is a template of instructions for loading an overlay. For each overlay routine in the program, the linker builds and stores a list of PLIT instances according to that template, as it builds its executable file. The linker may also save registers or stack context information. The linker does not accept a PLIT without arguments. If you do not want the linker to redirect function calls in overlays, omit the PLIT{} commands entirely.

To help write an overlay manager, the linker generates PLIT_SYMBOL constants for each symbol in an overlay. The overlay manager can also: • Be helped by manual intervention. Save the target’s state on the stack or in memory before loading and executing an overlay function, to ensure it continues correctly on return. However, you can implement this feature within the PLIT section of your .ldf file. Note: Your program may not need to save this information. • Initiate (jump to) the routine that transfers the overlay code to internal memory, after given the previous information about its identity, size, and location: _OverlayManager. “Smart” overlay managers first check whether an overlay function is already in internal memory to avoid reloading the function.

5-38 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

Linking Multiprocessor Systems

The linker has several commands that can be used to build executable images for multiprocessor systems. Selecting the right multiprocessor linking commands and using them depend on the system you are building and the Analog Devices processor in your system. The linker will only support linking for homogeneous multiprocessors (that is, the system must use the same kind of processor throughout). If you are building a heterogeneous multiprocessing environment, you will need to build the system with more than one link step, using an .ldf file for each kind of processor in your system.

A homogeneous multiprocessor system can be linked with a single .ldf file. The .ldf file will have a PROCESSOR{} command that describes which object files and libraries are to be linked into the memory for each processor. Every PROCESSOR{} command will produce a separate executable file (.dxe). For processors that can access the local memory of other processors (for example, through link ports), the MPMEMORY{} command can be used to define the offset of each processor's physical memory. The MPMEMORY{} command is described below. It is possible to specify the code and data that is to be placed into memory that is shared between processors. Two commands are available for placing objects and libraries into shared memory: SHARED_MEMORY{} and COMMON_MEMORY{}. Which of these commands you use will depend on how you intend to use the shared memory and the limitations of the processor architecture. The SHARED_MEMORY{} command can be used if the shared memory in the system does not contain any references to memory that is internal to an individual processor, or if the processor architecture supports addressing the internal memory of other processors. For other processors, such as ADSP-BF561 processors, where one processor can not access the internal memory of the other processor, use the

VisualDSP++ 5.0 Linker and Utilities Manual 5-39 Linking Multiprocessor Systems

COMMON_MEMORY{} command. These commands and their usage are described in more detail below. This section describes the following features and LDF commands: • “Selecting Code and Data for Placement” • “Mapping by Section Name” on page 5-42 • “Mapping Using Attributes” on page 5-43 • “Mapping Using Archives” on page 5-44 • “MPMEMORY{}” on page 5-45 • “SHARED_MEMORY{}” on page 5-47 • “COMMON_MEMORY{}” on page 5-53 Regardless of the linker commands that you use, you will have to make decisions regarding which code is going to run on which processor, where data will be placed, and what processors have access to what data. Once you have a partitioning of your code and data you can use the .ldf file to instruct the linker on code/data placement.

Selecting Code and Data for Placement

There are many ways to identify code and data objects for placement in a multiprocessor system. The methods are the same methods used when being selective about placement of objects in internal or external memory. There are advantages and disadvantages for each of the methods, and an .ldf file may combine many of these methods.

Using LDF Macros for Placement The easiest way to partition code and data between processors is to explicitly place the object files by name. In the example below, the code that is

5-40 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

to be placed in core A are in object files that are explicitly named in the .ldf file.

{ OUTPUT ( $COMMAND_LINE_OUTPUT_DIRECTORY/corea.dxe ) SECTIONS { code { INPUT_SECTIONS (corea.doj(program) coreamain.doj(program)) } > CoreaCode ... } PROCESSOR COREB { OUTPUT ( $COMMAND_LINE_OUTPUT_DIRECTORY/coreb.dxe ) SECTIONS { code { INPUT_SECTIONS (coreb.doj(program) corebmain.doj(program)) } > CorebCode ... }

Doing placement explicitly by object file can be made easier through the use of LDF macros. The example could be simplified with macros for the objects to be placed in each core.

$COREAOBJECTS = corea.doj, coreamain.doj; $COREBOBJECTS = coreb.doj, corebmain.doj; ...

VisualDSP++ 5.0 Linker and Utilities Manual 5-41 Linking Multiprocessor Systems

PROCESSOR COREA { ... SECTIONS { code { INPUT_SECTIONS ( $COREAOBJECTS(program) ) } > CoreaCode }

By using an LDF macro, it is much easier to make changes if functionality is going to be moved from one processor to another. Object files can appear in more than one LDF macro. Depending on the system, the same object file may be mapped to more than one processor. The main advantages of explicitly naming object files when placing object files to processors is that it is explicit in the .ldf file where each object file goes. By using LDF macros, the list of object files can be localized. A disadvantage for explicitly naming object files is that every time a new file is added to your system, the .ldf file must be modified to explicitly reference the file. Also, it is not possible to share the .ldf file with other projects that are built on the same multiprocessing system.

Mapping by Section Name

Both the compiler and assembler allow you to name sections in object files. In the assembler, this is done using the .SECTION directive: .SECTION Corea_Code;

The compiler has two ways to name a section. The first method uses the section() qualifier: section("Corea_Code") main() {...}

5-42 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

The section name can also be specified using the section pragma. The use of this pragma is recommended since it is more flexible and results in code that is portable.

#pragma section ("Corea_Code") main() {...}

Users can use section names to identify code that is to be placed with a particular processor.

PROCESSOR COREA { OUTPUT ( $COMMAND_LINE_OUTPUT_DIRECTORY/corea.dxe ) SECTIONS { code { INPUT_SECTIONS ( $OBJECTS(Corea_Code) ) } > CoreaCode ... }

The advantage of mapping by section name is that the .ldf file can be made generic and reused for other projects using the same multiprocessor. The disadvantage is that it requires making changes to C and assembly source code files to make the mapping. Also, it may not be possible to modify source code for some libraries or code supplied by third parties.

Mapping Using Attributes

The linker now supports mapping by attributes. When compiling and assembling, users can assign attributes to object files. These attributes can then be used to filter object files for inclusion (or exclusion) during mapping. Users can assign attributes to object files that identify a core that the object files should be mapped to, a core that an object file should not be mapped to, code that is safe to be shared by all processors, and so on.

VisualDSP++ 5.0 Linker and Utilities Manual 5-43 Linking Multiprocessor Systems

The run-time libraries are built using attributes so it possible to select areas within the run-time libraries for placement. For example, it is possible to select the objects in the run-time libraries that are needed for I/O and place them only in external memory.

An advantage of using attributes is that the .ldf file can be made generic and reused for other projects using the same multiprocessor. The disadvantage is that changing where an object is placed requires rebuilding the object file in order to change the attributes. Also, if all of the object files are being built in the same project, it can be inconvenient to use file-specific build options. Also, it may not be possible to rebuild the object for some libraries.

Mapping Using Archives

Another way to partition files is to build an object archive or library. As an example, you could create a project just for building the object files to be placed in core A. The target of the project would be an archive named corea.dlb. The project that actually links the multiprocessor system would include corea.dlb. In fact, it is easiest to build a project group in which the linking project would have dependencies on the projects that build the archives it depends on. The .ldf file would then use the archive for linking:

PROCESSOR COREA { OUTPUT ( $COMMAND_LINE_OUTPUT_DIRECTORY/corea.dxe ) SECTIONS { code { INPUT_SECTIONS ( corea.dlb(program) ) } > CoreaCode

5-44 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

... }

The disadvantage of using archives for mapping is that it requires organiz- ing more than one project. The advantage is that it can be easy to add, delete, or move objects from one processor to another. Removing an object from a project will remove it from the archive when the project is rebuilt. Adding a file to a project that builds an archive will automatically add the file to the link without needing to make changes to source. This flexibility makes it easy to create an .ldf file that can be shared by users building for the same architecture.

The COMMON_MEMORY{} command requires archives when mapping objects into memory that is shared between processors. This command is described in more detail in “COMMON_MEMORY{}” on page 5-53.

MPMEMORY{} L The MPMEMORY{} command is not used with Blackfin processors. The MPMEMORY{} command specifies the offset of each processor’s physical memory in a multiprocessor target system. After you declare the processor names and memory segment offsets with the MPMEMORY{} command, the linker uses the offsets during multiprocessor linking. Refer to “Memory Overlay Support” on page 5-8 for a detailed description of overlay functionality.

Your .ldf file (and other .ldf files that it includes), may contain one MPMEMORY{} command only. The maximum number of processors that you can declare is architecture-specific. Follow the MPMEMORY{} command with PROCESSOR processor_name{} commands, which contain each processor’s MEMORY{} and SECTIONS{} commands.

VisualDSP++ 5.0 Linker and Utilities Manual 5-45 Linking Multiprocessor Systems

Figure 5-10 shows MPMEMORY{} command syntax.

MPMEMORY{shared_segment_commands}

processor_name { START(address_expression) }

Figure 5-10. MPMEMORY{} Command Syntax Tree

Definitions for parts of the MPMEMORY{} command’s syntax are:

• shared_segment_commands – Contains processor_name declarations with a START{} address for each processor’s offset in multiprocessor memory. Processor names and linker labels follow the same rules. For more information, refer to “LDF Expressions” on page 3-20.

• processor_name{placement_commands} – Applies the processor_name offset for multiprocessor linking. Refer to “PROCESSOR{}” on page 3-54 for more information.

The MEMORY{} command specifies the memory map for the target L system. The .ldf file must contain a MEMORY{} command for global memory on the target system and may contain a MEMORY{} command that applies to each processor’s scope. An unlimited number of memory segments can be declared within each MEMORY{} command. For more information, see “MEMORY{}” on page 3-44. See “Memory Characteristics Overview” on page 2-27 for memory map descriptions.

5-46 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

SHARED_MEMORY{}

The SHARED_MEMORY{} command creates an executable output that maps code and data into a memory space that is shared by multiple processors. The output is given the extension .sm for shared memory. The SHARED_MEMORY{} command is similar in structure to the PROCESSOR{} command. The PROCESSOR{} command contains, among other commands, an OUTPUT() command that specifies a .dxe file for the output, and uses SECTIONS{} command to map selected sections from object files into specified sections in processor memory. Similarly, the SHARED_MEMORY{} command uses an OUTPUT() command and SECTIONS{} command to create an .sm file.

Figure 5-11 shows the syntax for the SHARED_MEMORY{} command, followed by definitions of its components.

SHARED_MEMORY { OUTPUT(file_name.SM) SECTIONS {section_commands} }

Figure 5-11. SHARED_MEMORY{} Command Syntax

The command components are:

• OUTPUT() – Specifies the output file name (file_name.sm) of the shared memory executable (.sm) file. An OUTPUT() command in a SHARED_MEMORY{} command must appear before the SECTIONS{} command in that scope.

• SECTIONS() – Defines sections for placement within the shared memory executable (.sm) file.

The .ldf file will have a MEMORY{} command that defines the memory configuration for the multiprocessor. The SHARED_MEMORY{} command must appear in the same LDF scope as the MEMORY{} command. The

VisualDSP++ 5.0 Linker and Utilities Manual 5-47 Linking Multiprocessor Systems

PROCESSOR{} commands for each processor in the system should also appear at this same LDF scope.

Figure 5-12 shows the scope of SHARED_MEMORY{} commands in the LDF.

MEMORY { my_shared_ram { TYPE(PM RAM) START(5120k) LENGTH(8k) WIDTH(32) } } SHARED_MEMORY {

OUTPUT(shared.sm )

SECTIONS { my_shared_sections{section_commands} > my_shared_ram } } PROCESSOR p0 { processor_commands with link against shared memory } PROCESSOR p1 { processor_commands with link against shared memory }

Figure 5-12. LDF Scopes for SHARED_MEMORY{}

The mapping of objects into processors and shared memory is made useful by being able to have processors and shared memory “link against” each other. The LINK_AGAINST() command specifies a .dxe file or .sm file generated by the mapping for another processor or shared memory and makes the symbols in that file available for resolution for the current processor.

The MEMORY{} command appears in a scope that is available to any SHARED_MEMORY{} command or PROCESSOR{} command that uses the shared memory. To achieve this type of scoping across multiple links, place the

5-48 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

shared MEMORY{} command in a separate .ldf file and use the INCLUDE() command to include that memory in both links.

When the .dxe file or .sm file that is named in the LINK_AGAINST() command is generated by another .ldf file, the linker will read in the executable file just as it reads in object files and archives. When the .dxe file of the .sm file that is named is being generated in the same .ldf file, the linker will use the executable file as it is being generated. When the processor and shared memory appear in the same .ldf file, the order that the processor or shared memory commands appear is not important.

For example, consider that the object file data.doj contains the global data buffer DBUF, and the object file main.doj contains code that references that data. Further, the data buffer DBUF is placed in shared memory so that it is available to multiple processors, while main.doj contains code that is going to be executed from core A. An .ldf file that does this mapping would include:

SHARED_MEMORY { OUTPUT("shared_memory.sm") SECTIONS { data_sm { INPUT_SECTIONS(data.doj(data)) } > mem_shared_mem } } PROCESSOR CoreA { OUTPUT("corea.dxe") LINK_AGAINST("shared_memory.sm") SECTIONS {

VisualDSP++ 5.0 Linker and Utilities Manual 5-49 Linking Multiprocessor Systems

code_corea { INPUT_SECTIONS(main.doj(program)) } > corea_a_mem } }

In the example .ldf file, the SHARED_MEMORY{} command creates the output file shared_memory.sm. The data from the object file data.doj is mapped into the output file and placed into the memory named mem_shared_mem. (The memory definition is not shown.) Later in the .ldf file. the mapping for core A is done with a PROCESSOR{} command. In addition to creating the output file (corea.dxe) and mapping the program sections from the object file main.doj, it also “links against” the file corea.dxe.

The LINK_AGAINST() command has the following effect: After all of the objects and sections for processor core A have been mapped, the symbol table in the file shared_memory.sm is used to find any symbols that could not be resolved. In the example, the object file main.doj contains a reference to the DBUF symbol but none of the object files mapped into core A contained that symbol. The symbols in shared_memory.sm are then read and DBUF is found to have been mapped into shared memory. The linker will resolve the reference in core A to be the address in shared memory that DBUF was mapped into by processing the SHARED_MEMORY{} command that produced shared_memory.sm. The processing order described above is slightly modified if there are symbols that have weak linkage. A symbol with strong linkage in an executable named in a LINK_AGAINST() command will take precedence over a “weak” symbol.

The LINK_AGAINST() command takes effect only after mapping of objects and libraries in the input sections for the processor. Object from libraries will be mapped if needed to resolve references, even if those symbols are available in the shared memory .sm file named in the LINK_AGAINST()

5-50 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

command. If the processor and shared memory both map the same library files, it is possible that an object from that library may get mapped into the processor and the shared memory. The multiple mapping is unlikely to make the program incorrect, but it can be a waste of memory.

The LINK_AGAINST() command can also appear within a SHARED_MEMORY{} command. It is possible for a shared memory to link against a processor .dxe file. The LINK_AGAINST() command works in the same way. After mapping objects and libraries that are listed in INPUT_SECTIONS() commands, if there are symbols that have not been resolved, the .dxe file (or .sm file) specified in the LINK_AGAINST() will be used.

It is possible for more than one LINK_AGAINST() command to appear in the same processor or shared memory. The .dxe files or .sm files that are named will be searched in the order they appear to resolve references. It is also possible to have a processor link against a shared memory and have the same shared memory link against that processor. The bidirectional link against can allow code in the processor memory to call code that exists in shared memory that can then call code that is in the processor memory. As mentioned above, linking behavior does not depend on the order that processors and shared memory appear in the .ldf file. This order independence is still true with a bidirectional link against. Note that references from shared memory into processor memory may not be supported by all processors. For example, for a multi-core Blackfin processor like the ADSP-BF561 processor, it is not possible for code executing in one core to access memory that is in internal memory of the other processor. If there is code in shared memory that references internal memory of core A, that code can only be executed on core A. If core B executes the code, once core B tries to reference the internal memory on core A, the part will halt because of a hardware exception.

VisualDSP++ 5.0 Linker and Utilities Manual 5-51 Linking Multiprocessor Systems

Also note that on parts where processors can access the internal memory of the other processors, that access may be slow and affect the performance of your program.

If you don't have LINK_AGAINST() commands within a SHARED_MEMORY{} command then there won't be any references from shared memory back to internal memory of any of the cores. If your system needs to have references from shared memory back to processors it is best to use the COMMON_MEMORY{} command. If there are references from shared memory back to processor internal memory for the Blackfin processors, COMMON_MEMORY{} is required. One solution is to partition shared memory into a section reserved for core A, a section reserved for core B, and a section that is memory shared between the two processors. The partitioning is managed by using the MEMORY{} command. Then the PROCESSOR{} command for core A will map into the core A internal memory and into the section of shared memory reserved for core A. It will also typically link against the shared memory. The PROCESSOR{} command for core B will map into the core B internal memory and into the section of shared memory reserved for core B, and link against the shared memory. The SHARED_MEMORY{}} command is used to map the program and data that is common to both processors.

5-52 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

COMMON_MEMORY{}

The COMMON_MEMORY{} command provides another way to map objects into memory that is shared by more than one processor. The mapping is done in the context of the processors that will use the shared memory; these processors are identified as a “master” of the common memory. The COMMON_MEMORY{} command will also manage references from the shared memory back to internal memory of the processors so that each processor will not reference memory that is in another processor’s internal memory. The COMMON_MEMORY{} command looks like the PROCESSOR{} and SHARED_MEMORY{} commands in that it uses INPUT_SECTIONS() commands for mapping. A restriction is that within a COMMON_MEMORY{} command, only archives may be mapped and not individual object files. The following example shows the basic components of the COMMON_MEMORY{} command.

COMMON_MEMORY { OUTPUT("common_memory.cm") MASTERS(CoreA, CoreB) SECTIONS { data_cm { INPUT_SECTIONS(common.dlb(data)) } > mem_common_mem } } PROCESSOR CoreA { OUTPUT("corea.dxe") SECTIONS {

VisualDSP++ 5.0 Linker and Utilities Manual 5-53 Linking Multiprocessor Systems

code_corea { INPUT_SECTIONS(main.doj(program)) } > corea_a_mem } } PROCESSOR CoreB { OUTPUT("coreb.dxe") SECTIONS { code_corea { INPUT_SECTIONS(main.doj(program)) } > corea_a_mem } }

The COMMON_MEMORY{} command uses the OUTPUT() to name the file that will hold the result of the mapping. The command uses the .cm extension for the file. The COMMON_MEMORY{} command also uses the SECTIONS{} command to map files into memory segments. However, the only files that can be mapped are archive (.dlb) files. Individual object files cannot be mapped from inside of a COMMON_MEMORY{} command.

The biggest syntactic difference in the COMMON_MEMORY{} command is the MASTERS() command. This command explicitly identifies the processors that are going to share the memory. The processor names are the name used in the PROCESSOR{} commands also appearing in the same .ldf file. Within the PROCESSOR{} command, there is no need for a LINK_AGAINST() command specifying the common memory. The MASTERS() command describes the connection.

The mapping of the archives in the COMMON_MEMORY{} command is really done when the mapping is done for the masters named in the MASTERS()

5-54 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

command. While mapping for each of the processors named as a master, the linker will treat each INPUT_SECTIONS() command in the common memory as if they appeared within the PROCESSOR{} command. Since only archives are allowed, only the objects within the archive that are needed to satisfy references for the processor will be mapped. The mapping will be into the memory sections in the common memory. For example, the effect of the previous example will be as if the INPUT_SECTIONS() in the COMMON_MEMORY{} were part of the PROCESSOR{}:

// NOT ACTUAL LDF – EFFECT OF COMMON_MEMORY{} PROCESSOR CoreA { OUTPUT("corea.dxe") SECTIONS { code_corea { INPUT_SECTIONS(main.doj(program)) } > corea_a_mem // when mapping CoreA, the input sections from // the common memory are mapped as if they were // part of this PROCESSOR{} because CoreA is // listed as a MASTER data_cm { INPUT_SECTIONS(common.dlb(data)) } > mem_common_mem } }

Of course, by specifying with the COMMON_MEMORY{} command, the same mapping for the objects in common.dlb will also be done for core B, and the objects that are shared by the two processors will only be mapped once into the shared memory space.

VisualDSP++ 5.0 Linker and Utilities Manual 5-55 Linking Multiprocessor Systems

The mapping will be done for each of the processors named as a master. Some symbols will be needed for each processor, and in simple cases the common memory will share the code or data between the processors. If an object is mapped into common memory that has a reference that goes back into internal memory of a processor, if necessary, the linker will make a copy of the object file so that both cores can safely use common memory. This behavior is described in the example below. To demonstrate the complexities of multiprocessing linking, the example has several dependencies. The abbreviated C examples show the dependencies for several object files.

// file mainA.doj void mainA() { // the main code in CoreA references 2 common functions commonfunc1(); commonfunc2(); } // file mainB.doj void mainB() { // the main code in CoreB references 3 common functions commonfunc1(); commonfunc2(); commonfunc3(); } // file func1.doj void commonfunc1() { // a common function with a reference to a library libfunc1(); } // file func2.doj void commonfunc2() { // a common function with a reference to a library libfunc2(); }

5-56 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

// file func3.doj void commonfunc3() { // no further references } // file libfunc1.doj and libfunc2.doj have no further references // create archives for common files elfar –c common.dlb func1.doj func2.doj func3.doj elfar –c commonlib.dlb libfunc1.doj libfunc2.doj

Each of the processors has its own main function. Each main function makes calls to common functions. Some of the common functions make further calls to library functions. The common functions have been placed in an archive named common.dlb, and the library files have been placed in an archive named commonlib.dlb.

The .ldf file to build the multiprocessor system is shown below.

COMMON_MEMORY { OUTPUT("common_memory.cm") MASTERS(CoreA, CoreB) SECTIONS { data_cm { // the common libraries are mapped into common // memory INPUT_SECTIONS(common.dlb(program) commonlib.dlb(program)) } > mem_common_mem } } PROCESSOR CoreA { OUTPUT("corea.dxe")

VisualDSP++ 5.0 Linker and Utilities Manual 5-57 Linking Multiprocessor Systems

SECTIONS { code_corea { INPUT_SECTIONS(mainA.doj(program)) // for performance reasons map // libfunc1.doj into this core INPUT_SECTIONS(libfunc1.doj(program)) } > corea_a_mem } } PROCESSOR CoreB { OUTPUT("coreb.dxe") SECTIONS { code_coreb { INPUT_SECTIONS(mainB.doj(program)) } > corea_b_mem } }

Notice that processor core A explicitly maps libfunc1.doj into its internal memory. Core B does not map a version of libfunc1.doj. Both processors link against the common memory that does mapping against the archives that contain common functions.

To understand the operation of COMMON_MEMORY{}, let’s walk through the mapping of the objects into memory, beginning with core A. The INPUT_SECTIONS() commands for core A will map mainA.doj and libfunc1.doj into the memory corea_a_mem. The references to commonfunc1 and commonfunc2 will cause the object files func1.doj and func2.doj to be pulled out of the archive common.dlb and they will be mapped into the common memory mem_common_mem. The object file

5-58 VisualDSP++ 5.0 Linker and Utilities Manual Memory Overlays and Advanced LDF Commands

func1.doj has a reference to libfunc1. This symbol was already mapped when libfunc1.doj was mapped into the core memory. The object file func2.doj has a reference to libfunc2 so the object libfunc2.doj will be pulled out of the archive commonlib.dlb and it will also be mapped into mem_common_mem. Note that this mapping only considers the files required for core A so commonfunc3 is not considered.

The mapping for core B will be similar. The INPUT_SECTIONS() command for core B will map mainB.doj into the memory coreb_b_mem. The references to the common functions will cause the object files func1.doj, func2.doj, and func3.doj to be pulled out of the archive common.dlb and be mapped into mem_common_mem. The references in the common functions to the library functions will cause the library objects to be pulled from the commonlib.dlb so libfunc1.doj and libfunc2.doj will be mapped into the common memory mem_common_mem. Note that this mapping only considers the files for core B and the common memory. In particular, the fact that libfunc1.doj was mapped into core A memory is not considered for this mapping. Now the linker ensures that all the objects mapped into common memory can be shared; for those files that cannot be shared, it will fix them by making duplications. Those object files mapped into common memory that do not have any further references (the leaf functions func3.doj, libfunc1.doj, and libfunc2.doj) are fine as they are. The function commonfunc2 references libfunc2.doj (which is only mapped into common memory), so it is also fine. The function commonfunc1 references libfunc1.doj. In the context of core A, func1.doj will call the version of libfunc1 that is mapped into core A internal memory. In the context of core B, func1.doj will call the version of libfunc1 that is mapped into common memory. To resolve this problem, the linker will create a copy of func1.doj. The mainA function will call the version that references back to the version of libfunc1 that is in core A memory while mainB will call the version that references back to the version of libfunc1 that is in common memory.

VisualDSP++ 5.0 Linker and Utilities Manual 5-59 Linking Multiprocessor Systems

It is rare that an object mapped into common memory will be duplicated. When an object is duplicated, the linker will only duplicate the minimal amount needed to keep integrity. The duplication will only happen in cases where using the SHARED_MEMORY{} command would have resulted in a run-time exception, because a processor was accessing memory in another processor’s internal memory.

5-60 VisualDSP++ 5.0 Linker and Utilities Manual 6ARCHIVER

The VisualDSP++ archiver (elfar) combines object (.doj) files into library files, which serve as reusable resources for code development. The VisualDSP++ linker rapidly searches library files for routines (library members) referred to by other object files and links these routines into the executable program. This chapter provides: • “Introduction” on page 6-2 Introduces the archiver’s functions • “Archiver Guide” on page 6-3 Describes the archiver’s functions • “Archiver Command-Line Reference” on page 6-14 Describes archiver operations by means of command-line switches

VisualDSP++ 5.0 Linker and Utilities Manual 6-1 Introduction

Introduction

The elfar utility combines and indexes object files (or any other files) to produce a searchable library file. It performs the following operations, as directed by options on the elfar command line: • Creates a library file from a list of object files • Appends one or more object files to an existing library file • Deletes file(s) from a library file • Extracts file(s) from a library file • Prints the contents of object files of an existing library file to stdout

• Replaces file(s) in an existing library file • Encrypts symbol(s) in an existing library file

• Embeds version information into a library built with elfar The archiver can run only one of these operations at a time. However, for commands that take a list of file names as arguments, the archiver can input a text file that contains the names of object files (separated by white space). The operation makes long lists easily manageable. The archiver, sometimes called a librarian, is a general-purpose utility. It combines and extracts arbitrary files. This manual refers to DSP object (.doj) files because they are relevant to DSP code development.

6-2 VisualDSP++ 5.0 Linker and Utilities Manual Archiver

Archiver Guide

The elfar utility combines and indexes object files (or any other files) to produce a searchable library file. This section describes the following archiver functions: • “Creating a Library” • “Making Archived Functions Usable” on page 6-4 • “Archiver Symbol Name Encryption” on page 6-10

Creating a Library

To create an archive, use the -c switch when invoking the archiver from the command line (as shown in “Archiver Command-Line Reference” on page 6-14). The command line should include the name of the archive being created and the list of objects files to be added. Example: elfar -c my_lib.dlb fft.doj sin.doj cos.doj tan.doj

If the objects files were created using the C/C++ compiler, it is recommended that the compiler driver and the compiler’s -build-lib switch are used to build the library (the compiler driver invokes elfar to build the library). Refer to the appropriate VisualDSP++ 5.0 C/C++ Compiler and Library Manual for more information. Example: ccblkfn -build-lib -o my_lib.dlb fft.doj sin.doj cos.doj tan.doj

On Window systems, it is possible to build a library from within the VisualDSP++ development environment. VisualDSP++ writes its output to .dlb.

VisualDSP++ 5.0 Linker and Utilities Manual 6-3 Archiver Guide

To maintain code consistency, use the conventions in Table 6-1.

Table 6-1. File Name Extensions used with Archiver

Extension File Description

.dlb Library file

.doj Object file. Input to archiver.

.txt Text file used as input with the -i switch

Making Archived Functions Usable

In order to use the archiver effectively, you must know how to write archive files, which make your DSP functions available to your code (via the linker), and how to write code that accesses these archives. Archive usage consists of two tasks: • Creating library routines, functions that can be called from other programs, and library data, variables, that can be referenced from programs • Accessing library routines and data from your code

Writing Archive Routines: Creating Entry Points A library routine (or function) in code can be accessed by other programs. Each routine must have a globally visible start label (entry point). Library data must be given a visible label. Code that accesses that routine must declare the entry point’s name as an external symbol in the calling code. To create visible external symbol: 1. Declare the start label of each routine and each variable as a global symbol with the assembler’s .GLOBAL directive. This defines the entry point.

6-4 VisualDSP++ 5.0 Linker and Utilities Manual Archiver

The following code fragment has a visible entry point for the function dIriir and creates a visible symbol for the variable FAE. ... .global dIriir; .section data1; .byte2 FAE = 0x1234,0x4321;

.section program; .global FAE; dIriir: R0=N-2; P2 = FAE;

2. Assemble the files into object files containing the global segments. 3. You can also write library functions in C and C++. Functions declared in your C/C++ file will be given globally visible symbols that can be referenced by other programs. Use the C/C++ compiler to create objects files, and use the compiler driver and its -build-lib switch to create the library.

Accessing Archived Functions From Your Code

Programs that call a library routine must use the assembler’s .EXTERN directive to specify the routine’s start label as an external label. When linking the program, specify one or more library (.dlb) files to the linker, along with the names of the object (.doj) files to link. The linker then searches the library files to resolve symbols and links the appropriate routines into the executable file. Any file containing a label referenced by your program is linked into the executable output file. Linking libraries is faster than using individual object files, and you do not have to enter all the file names, just the library name.

VisualDSP++ 5.0 Linker and Utilities Manual 6-5 Archiver Guide

In the following example, the archiver creates the filter.dlb library containing the object files: taps.doj, coeffs.doj, and go_input.doj. elfar -c filter.dlb taps.doj coeffs.doj go_input.doj

If you then run the linker with the following command line, the linker links the object files main.doj and sum.doj, uses the default .ldf file (for example, ADSP-BF533.ldf), and creates the executable file (main.dxe). linker -DADSP-BF533 main.doj sum.doj filter.dlb -o main.dxe

Assuming that one or more library routines from filter.dlb are called from one or more of the object files, the linker searches the library, extracts the required routines, and links the routines into the .dxe file.

Specifying Object Files The list of object files on the command line is used to specify objects to be added to the archive. Such commands are -c (create), -a (add), or -r (replace). The list can also be used to specify objects in the library to be extracted using the -e (extract) command. When the list refers to object files to be added to the archive, the file name is specified the way the file names are specified for the host operating system. The file name can include path information – relative or absolute. If path information is not included, the archiver will look for the file in the current working directory. When the list refers to object files already in the archive, the file names should not include any path information. The archiver only saves the base file name for the object files in the archive.

The archiver accepts the wildcard character “*” in the specification of the object file names. On Windows systems, the archiver does all interpreta- tion of the wildcard character. When it appears in a list of object files to be added, the archiver searches the file system for files that match this specification. When a wildcard appears in a list of objects already in the

6-6 VisualDSP++ 5.0 Linker and Utilities Manual Archiver

library, the archiver will search through the object files in the library for matches.

Tagging an Archive With Version Information The archiver supports embedding version information into a library built with elfar.

Basic Version Information You can “tag” an archive with a version. The easiest way to tag an archive is with the -t switch (see Table 6-2 on page 6-15), which takes an argument (the version number). For example,

elfar –t 1.2.3 lib.dlb

The –t switch can be used in addition to any other elfar command. For example, a version can be assigned at the same time that a library is created:

elfar –c –t "Steve's sandbox Rev 1" lib.dlb *.doj

To hold version information, the archiver creates an object file, __version.doj, that has version information in the .strtab section. This file is not made visible to the user.

An archive without version information will not have the __version.doj entry. The only operations on the archive using elfar that add version information are those that use the –t switch. That is, an archive without version information does not pick up version information unless specifically requested.

If an archive contains version information (__version.doj is present), all operations on the archive preserve that version information, except operations that explicitly request version information to be stripped from the archive (see “Removing Version Information From an Archive” on page 6-10).

VisualDSP++ 5.0 Linker and Utilities Manual 6-7 Archiver Guide

If an archive contains version information, that information can be printed with the -p command.

elfar –p lib.dlb ::User Archive Version Info: Steve's sandbox Rev 1 a.doj b.doj

The archiver adds “::” to the front of the version information to highlight it.

User-Defined Version Information You can provide any number of user-defined version values by supplying a text file that contains those values. The text file can have any number of entries. Each line in the file begins with a name (a single token with no embedded white space), followed by a space and then the value associated with that name. As an example, consider the file foo.txt:

my_name neo my_location zion CVS_TAG matrix_v_8_0 other version value can be many words; name is only one

This file defines four version names: my_name, my_location, CVS_TAG, and other. The value of my_name is neo; the value of other is “version value can be many words; name is only one”.

To tag an archive with version information from a file, use the –tx switch (see Table 6-2 on page 6-15) which accepts the name of that file as an argument:

elfar –c –tx foo.txt lib.dlb object.doj elfar –p lib.dlb ::CVS_TAG matrix_v_8_0 ::my_location zion ::my_name neo

6-8 VisualDSP++ 5.0 Linker and Utilities Manual Archiver

::other version value can be many words; name is only one object.doj

Version information can be added to an archive that already has version information. The effect is additive. Version information already in the archive is carried forward. Version information that is given new values is assigned the new values. New version information is added to the archive without destroying existing information.

Printing Version Information

As mentioned above, when printing the contents of an archive, the –p command (see Table 6-2 on page 6-15) prints any version information. Two forms of the –p switch can be used to examine version information.

The –pv switch prints version information only, and does not print the contents of the archive. This switch provides a quick way to check the version of an archive.

The –pva switch prints all version information. Version names without values cannot not be printed with –p or –pv but are shown with –pva. In addition, the archiver keeps two additional kinds of information:

elfar –a lib.dlb t*.doj elfar –pva lib.dlb ::User Archive Version Info: 1.2.3 ::elfar Version: 4.5.0.2 ::__log: -a lib.dlb t*.doj

The archiver version that created the archive is stored in __version.doj and is available using the –pva switch. Also, if any operations that cause the archive to be written were executed since adding version information, these commands appear as part of special version information called “__log”. The log prints a line for every command that has been done on the archive since version information was added to the archive.

VisualDSP++ 5.0 Linker and Utilities Manual 6-9 Archiver Guide

Removing Version Information From an Archive Every operation has a special form of switch that can cause an archive to be written and request that the version information is not written to the archive. Version information already in the archive would be lost. Adding “nv” (no version) to a command strips version information. For example,

elfar –anv lib.dlb new.doj

elfar –dnv lib.dlb *

In addition, a special form of the –t switch (see Table 6-2 on page 6-15), which takes no argument, can be used for stripping version information from an archive:

elfar –tnv lib.dlb // only effect is to remove version info

Checking Version Number You can have version numbers conform to a strict format. The archiver confirms that version numbers given on the command line conform to an nn.nn.nn format (three numbers separated by “.”). The –twc switch (see Table 6-2 on page 6-15) causes the archiver to raise a warning if the version number is not in this form. The check ensures that the version number starts with a number in this format. For example,

elfar –twc "1.2 new library" lib.dlb [Warning ar0081] Version number does not match num.num.num format Version 0.0.0 will be used. elfar –pv lib.dlb ::User Archive Version Info: 0.0.0 1.2 new library

Archiver Symbol Name Encryption

Symbol name encryption protects intellectual property contained in an archive (.dlb) library that might be revealed when using meaningful symbol names. Code and test a library with meaningful symbol names,

6-10 VisualDSP++ 5.0 Linker and Utilities Manual Archiver

and then use archive library encryption on the fully tested library to disguise the names. Source file names in the symbol tables of object files in the archive L are not encrypted. The encryption algorithm is not reversible. Also, encryption does not guarantee a given symbol is encrypted the same way when different libraries, or different builds of the same library, are encrypted.

The -s switch (see Table 6-2) is used to encrypt symbols in to produce . Symbols in are not encrypted, and provides the first letter of scrambled names. Command Syntax The following command line encrypts symbols in an existing archive file. elfar -s [-v] library_file in_library_file exclude_file type-letter

where:

-s – Selects the encryption operation.

-v – Selects verbose mode, which provides statistics on the encrypted symbols.

library_file – Specifies the name of the library (.dlb) file to be produced by the encryption process

in_library_file – Specifies the name of the archive (.dlb) file to be encrypted. This file is not altered by the encryption process, unless in-archive is the same as out-archive.

VisualDSP++ 5.0 Linker and Utilities Manual 6-11 Archiver Guide

exclude-file – Specifies the name of a text file containing a list of symbols not to be encrypted. The symbols are listed one or more to a line, separated by white space.

type-letter – The initial letter of type-letter provides the initial letter of all encrypted symbols. Encryption Constraints All local symbols can be encrypted, unless they are correlated with a symbol having external binding that should not be encrypted. Symbols with external binding can be encrypted when they are used only within the library in which they are defined. Symbols with external binding that are not defined in the library (or are defined in the library and referred to outside of the library) should not be encrypted. Symbols that should not be encrypted must be placed in a text file, and the name of that file given as the exclude-file command-line argument. Some symbol names have a prefix or suffix that has special meaning. The debugger does not show a symbol starting with “.” (period), and a symbol starting with “.” and ending with “.end” is correlated with another symbol. For example, “.bar” would not be shown by the debugger, and “._foo.end” would correlated with the symbol “_foo” appearing in the same object file. The encryption process encrypts only the part of the symbol after any initial “.” and before any final “.end”. This part is called the root of the symbol name. Since only the root is encrypted, a name with a prefix or suffix having special meaning retains that special meaning after encryption. The encryption process ensures that a symbol with external binding is encrypted the same way in all object files contained in the library. This process also ensures that correlated symbols within an object file are encrypted the same way, so they remain correlated.

The names listed in the exclude-file are interpreted as root names. Thus, “_foo” in the exclude-file prevents the encryption of the symbol names “_foo”, “._foo”, “_foo.end”, and “._foo.end”.

6-12 VisualDSP++ 5.0 Linker and Utilities Manual Archiver

The type-letter argument, which provides the first letter of the encrypted part of a symbol name, ensures that the encrypted names in different archive libraries can be made distinct. If different libraries are encrypted with the same type-letter argument, unrelated external symbols of the same length may be encrypted identically.

VisualDSP++ 5.0 Linker and Utilities Manual 6-13 Archiver Command-Line Reference

Archiver Command-Line Reference

The archiver processes object files into a library file with a .dlb extension, which is the default extension for library files. The archiver can also append, delete, extract, or replace member files in a library, as well as list them to stdout. This section provides the following reference information on the archiver command line and linking. • “elfar Command Syntax” • “Archiver Parameters and Switches” • “Command-Line Constraints” elfar Command Syntax

Use the following syntax to run elfar from the command line. elfar -[a|c|d|e|p|r] library_file object_file ...

Table 6-2 describes each switch. Example: elfar -v -c my_lib.dlb fft.doj sin.doj cos.doj tan.doj

This command line runs the archiver as follows:

-v – Outputs status information

-c my_lib.dlb – Creates a library file named my_lib.dlb

fft.doj sin.doj cos.doj tan.doj – Places these object files in the library file Table 6-1 on page 6-4 lists typical file types, file names, and extensions.

6-14 VisualDSP++ 5.0 Linker and Utilities Manual Archiver

Symbol Encryption

When employing symbol encryption, use the following syntax. elfar -s [-v] library_file in_library_file exclude_file type-letter

Refer to “Archiver Symbol Name Encryption” on page 6-10 for more information.

Archiver Parameters and Switches

Table 6-2 describes each archiver part of the elfar command. Switches must appear before the name of the archive file.

Table 6-2. Command-Line Switches and Entries

Item Description

exclude_file Specifies the name of a text file containing a list of symbols not to be encrypted.

lib_file Specifies the library that the archiver modifies. This parameter appears after the switch.

obj_file Identifies one or more object files that the archiver uses when modifying the library. This parameter must appear after lib_file. Use the -i switch to input a list of object files.

type-letter The initial letter of type-letter provides the initial letter of all encrypted symbols.

-a Appends one or more object files to the end of the specified library file

-anv Appends one or more object files and clears version information

-c Creates a new lib_file containing the listed object files

-d Removes the listed object files from the specified lib_file

-dnv Removes the listed obj_file(s) from the specified lib_file and clears version information

-e Extracts the specified file(s) from the library

VisualDSP++ 5.0 Linker and Utilities Manual 6-15 Archiver Command-Line Reference

Table 6-2. Command-Line Switches and Entries (Cont’d)

Item Description

-i filename Uses filename, a list of object files, as input. This file lists obj_file(s) to add or modify in the specified lib_file (.dlb).

-M Prints dependencies. Available only with the -c switch.

-MM Prints dependencies and creates the library. Available only with the -c switch.

-p Prints a list of the obj_file(s) (.doj) in the selected lib_file (.dlb) to standard output

-pv Prints only version information in library to standard output

-pva Prints all version information in library to standard output

-r Replaces the specified object file in the specified library file. The object file in the library and the replacement object file must have identical names.

-s Specifies symbol name encryption. Refer to “Archiver Symbol Name Encryption” on page 6-10.

-t verno Tags the library with version information in string

-tx filename Tags the library with full version information in the file

-twc ver Tags the library with version information in the num.num.num form

-tnv Clears version information from a library

-v (Verbose) Outputs status information as the archiver processes files

-version Prints the archiver (elfar) version to standard output

-w Disables archiver-generated warnings

-Wnnnn Selectively disables warnings specified by one or more message numbers. For example, -W0023 disables warning message ar0023.

The elfar utility enables you to specify files in an archive by using the wildcard character ‘*’. For example, the following commands are valid:

elfar –c lib.dlb *.doj // create using every .doj file

elfar –a lib.dlb s*.doj // add objects starting with 's'

elfar –p lib.dlb *1* // print files with '1' in their name

6-16 VisualDSP++ 5.0 Linker and Utilities Manual Archiver

elfar –e lib.dlb * // extract all files from the archive

elfar –d lib.dlb t*.doj // delete .doj files starting with 't'

elfar –r lib.dlb *.doj // replace all .doj files

The –c, -a, and –r switches use the wildcard to look up the file names in the file system. The -p, -e, and -d switches use the wildcard to match file names in the archive.

Command-Line Constraints

The elfar command is subject to the following constraints.

• Select one action switch (a, c, d, e, p, r, or s) only in a single command.

• Do not place the verbose operation switch, -v, in a position where it can be mistaken for an object file. It may not follow the lib_file during an append or create operation.

• The file include switch, -i, must immediately precede the name of the file to be included. The archiver’s -i switch enters a list of members from a text file instead of listing each member on the command line.

• Use the library file name first, following the switches. The -i and -v switches are not operational switches, and can appear later.

• When using the archiver’s -p switch, it is not necessary to identify members on the command line. • Enclose file names containing white space or colons within straight quotes. • Append the appropriate file extension to each file. The archiver assumes nothing, and does not do it for you.

VisualDSP++ 5.0 Linker and Utilities Manual 6-17 Archiver Command-Line Reference

• Wildcard options are supported with the use of the wildcard character “*”).

• The obj_file name (.doj object file) can be added, removed, or replaced in the lib_file. • The archiver’s command line is not case sensitive.

6-18 VisualDSP++ 5.0 Linker and Utilities Manual 7 MEMORY INITIALIZER

VisualDSP++ 5.0 supports the memory initializer tool. The memory initializer’s main function is to modify executable files (.dxe files) so that the programs are self-initializing. It does this by converting the program’s RAM-based contents into an initialization stream which it embeds into the executable file. This chapter provides: • “Memory Initializer Overview” on page 7-2 • “Basic Operation of Memory Initializer” on page 7-3 • “Initialization Stream Structure” on page 7-5 • “Run-Time Library Routine Basic Operation” on page 7-6 • “Using Memory Initializer” on page 7-7 • “Memory Initializer Command-Line Switches” on page 7-14

VisualDSP++ 5.0 Linker and Utilities Manual 7-1 Memory Initializer Overview

Memory Initializer Overview

The memory initializer may be used with processor systems where the RAM memory needs to be initialized with the code and data stored in the ROM memory before the execution of the application code begins. This is generally true for a processor system running in NO-BOOT mode. The initialization stream generated by the memory initializer is consumed by a dedicated run-time library (RTL) routine. Following a system reset, the RTL routine searches the initialization stream and initializes the processor’s RAM memory with the data in the initialization stream before the call to main(), the starting point of the application code. In creating the initialization stream, the memory initializer can, in most cases, effectively reduce the overall size of an executable file by combining contiguous, identical initialization into a single block. For example, a large zero-initialized array in an executable file can be compressed to a single small data block by the memory initializer.

In addition to a primary executable file (.dxe), the memory initializer accepts one or more additional executable files called “callback” executable files, and includes their data and instructions in the initialization stream. The RTL routine is able to call and execute them before conducting the process of the memory initialization for the primary application. This allows you to perform memory configuration and any other set-up functions that must occur before the code and data are extracted from ROM memory.

7-2 VisualDSP++ 5.0 Linker and Utilities Manual Memory Initializer

Basic Operation of Memory Initializer

This section describes the basic operations of the memory initializer, its input and output files, as well as basic initialization stream generated by the memory initializer.

Input and Output Files

The memory initializer takes an executable file (.dxe) as a primary input file and augments it by adding an initialization stream. The enhanced executable file is written as the output file. Processing the Primary Input Executable File After opening an input primary executable file, the memory initializer looks for sections, marked with the initialization flag in their section headers or specified from the command line, and extracts the data and instructions from them to make the primary initialization stream. By default, the stream is saved in the dedicated memory section called “.meminit” in the output file. For the sections from which the memory initializer extracts no data, the memory initializer simply copies them from the input file to the output file. Sections that are processed by the memory initializer to form the initialization stream are not needed in the output executable file, as their contents will be regenerated at runtime when the initialization stream is processed. Therefore, by default, such sections are not copied to the output file in order to reduce the size of the executable file. Processing Callback Input Executable Files In addition to a primary input executable file, the memory initializer optionally accepts a number of individually-built “callback” executable files specified with the -init switch (on page 7-16).

VisualDSP++ 5.0 Linker and Utilities Manual 7-3 Basic Operation of Memory Initializer

The memory initializer sequentially processes the callback executable files, one at a time. After opening an input callback executable file, the memory initializer looks for all of the sections marked with the initialization flag and PROGBITS qualifier (it indicates that the section contains instructions, data, or both), and extracts the data and instructions from them to make a callback initialization stream. When this stream is built up, the callback .dxe files are processed in the order specified on the command line. The memory initializer continues making a callback initialization stream from each of the callback executable files and pre-pending it to the primary initialization stream in the same sequence the callback executable files appear in the command line until the last callback executable file is processed. When processing a callback executable file, the memory initializer extracts all the code and data from it to make up the callback initialization stream regardless of the memory initializer command-line switches used only for the primary input file. Those switches are: • “-BeginInit Initsymbol” on page 7-15 • “-Init Initcode.dxe” on page 7-16 • “-NoAuto” on page 7-17 • “-NoErase” on page 7-17 • “-Section Sectionname” on page 7-18 This ensures the integrity of the code and data from each callback executable file in the callback initialization stream – the code can be executed independently and successfully, regardless of memory initializer command-line switches. By taking multiple input files, the memory initializer supports systems that have to run a number of independent service applications before starting the primary application.

7-4 VisualDSP++ 5.0 Linker and Utilities Manual Memory Initializer

Initialization Stream Structure

An initialization stream made from the memory initializer has three major portions: • The header of the initialization stream, which holds basic information for the run-time library (RTL) routine, such as the number of data blocks in the initialization stream • The callback executable file, which itself may have a number of the sub-portions, each containing a piece of the callback executable • The initialization data and code from the primary application Figure 7-1 shows the basic structure of the initialization stream:

Initialization Stream Header

First Callback Code (optional)

Second Callback Code (optional)

Additional Callback Code (optional)

Code and Data for the Primary Executable

Figure 7-1. Memory Initializer Basic Initialization Stream Structure

VisualDSP++ 5.0 Linker and Utilities Manual 7-5 Run-Time Library Routine Basic Operation

Run-Time Library Routine Basic Operation

A run-time library (RTL) routine performs the memory initialization with the initialization stream created by the memory initializer during runtime. It can be a dedicated RTL routine or user-provided routine called _mi_initialize (from the assembly code). For more information on the definition of the initialization stream, L see EE-239 for Blackfin processors. Following a system reset, the RTL routine is invoked by the application’s start-up code. The RTL routine: 1. Searches for the initialization stream 2. Digests the stream header 3. For each callback executable specified, copies “callback” code into RAM and then executes it. This is performed piece-by-piece and continues until execution is complete. 4. Brings the code and data from the primary executable file into the processor’s memory Once each callback executable has been executed, it is no longer needed in RAM; it may be overwritten by future callback executables or by the code or data spaces of the primary executable. After all the “callback” codes are executed, the RTL routine starts to initialize the processor’s memory with the initialization stream created from the primary input executable file, and overwrites the memory spaces previously initialized with the “callback” codes. After that, the RTL routine returns execution to the start-up header, and the application proceeds as normal.

7-6 VisualDSP++ 5.0 Linker and Utilities Manual Memory Initializer

If there are no callback executables to be executed, the RTL routine immediately starts the process of initializing memory for the primary application.

Using Memory Initializer

There are several reasons why it may be beneficial to use the memory initializer: • The system needs to initialize RAM memory from data stored in ROM. • It is desirable to reduce the overall size of the executable. • Initialization executable files need to run to configure the system, before the primary application starts. If it is decided to use the memory initializer, the preparation starts from the linker description file (.ldf) and the source files of the project.

Preparing the Linker Description File (.ldf)

If a section is to be processed by the memory initializer in order to create the initialization stream, the section must be marked in the .ldf file to indicate the kind of initialization required. This is done using initialization qualifiers (ZERO_INIT and RUNTIME_INIT). Sections marked with ZERO_INIT may contain only “zero-initialized” data, and sections marked with RUNTIME_INIT may contain the data with any initialization values.

Refer to the SECTIONS description (on page 3-61) for detailed L information on these qualifiers.

VisualDSP++ 5.0 Linker and Utilities Manual 7-7 Using Memory Initializer

The following example shows how to use the ZERO_INIT and RUNTIME_INIT qualifiers in an .ldf file to set up the section type.

my_zero_section ZERO_INIT { INPUT_SECTION_ALIGN(4) INPUT_SECTIONS( $OBJECTS(my_zero_section) $LIBRARIES(my_zero_section)) } >MEM_L1_DATA_A

my_data_section RUNTIME_INIT { INPUT_SECTION_ALIGN(4) INPUT_SECTIONS( $OBJECTS(my_data_section) ) }>MEM_L1_DATA_A

The section my_zero_section is intended to hold all the zero-initialized data, and the section my_data_section is to hold any other initialized data. After the program is first linked, the sections in the .dxe file have flags set according to the qualifiers in the .ldf file. Then the memory initializer runs and processes the .dxe file sections according to those flags, and produces a modified output .dxe file.

The memory initializer is able to identify the .dxe file sections with the distinct initialization flag and extract the data from them to make an initialization stream. Any number of sections can be set as either ZERO_INIT or RUNTIME_INIT type in an .ldf file.

Note that two memory sections are specified in default .ldf files, which also serve the memory initializer: bsz_init and .meminit. The bsz_init section holds the pointer generated by the memory initializer, which points to the start address of the initialization stream, and the section .meminit holds the actual initialization stream generated by the memory initializer. Although other sections may be selected as alternatives (using the appropriate command-line switches), this is not recommended.

7-8 VisualDSP++ 5.0 Linker and Utilities Manual Memory Initializer

Preparing the Source Files

The sections marked with the ZERO_INIT and RUNTIME_INIT qualifiers must be initialized with the proper values in the source files before being compiled. The following example shows one way to initialize a section.

#include #pragma section("my_data_section", RUNTIME_INIT) unsigned int A [ 100 ] = { 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd, 0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd,0xaabbccdd };

#pragma section("my_zero_section", ZERO_INIT) unsigned int B [ 128 ];

int main() {

VisualDSP++ 5.0 Linker and Utilities Manual 7-9 Using Memory Initializer

int i; int not_init = 0, not_zero = 0;

for (i = 0; i < 100; i++) if ( A [ i ] != 0xaabbccdd ) not_init++;

for (i = 0; i < 128; i++) if ( B [ i ] != 0 ) not_zero++;

printf ("A[]: %d elements not initialized/n", not_init); printf ("B[]: %d elements not zeroed/n", not_zero); return 0; }

Invoking Memory Initializer

There are several ways to invoke the memory initializer, either from the IDDE or from a command line.

Invoking meminit from the VisualDSP++ IDDE From the Project menu in the VisualDSP++ IDDE, chose Project Options, and select the Link page (see Figure 7-2). Type -meminit in the

7-10 VisualDSP++ 5.0 Linker and Utilities Manual Memory Initializer

Additional options field and then click OK. When the project is built, the linker calls the memory initializer.

Figure 7-2. Invoking the Memory Initializer from the VisualDSP++ IDDE

Invoking meminit from the Command Line The simplest command line to invoke the memory initializer is: meminit.exe input.dxe -o output.dxe

The memory initializer identifies all the sections with initialization flags in the input file, produces an initialization stream, and places it in the output file. Memory initializer command-line switches are listed in Table 7-1.

VisualDSP++ 5.0 Linker and Utilities Manual 7-11 Using Memory Initializer

Users of SHARC processors that have been using mem21k to invoke L the memory initializer from a command line can continue to do so. However, invoking meminit accomplishes the same results, since meminit passes the command to mem21k when used with a SHARC processor.

Invoking meminit from the Linker’s Command Line The simplest way to invoke the memory initializer from the linker’s command line is to use the linker’s -meminit switch. The linker also provides the -flag-meminit switch that passes each comma-separated option to the memory initializer. For example,

linker -proc ADSP-BF535 main.doj -meminit -o project1.dxe

Invoking meminit from the Compiler’s Command Line The simplest command line to invoke the memory initializer from the compiler’s command line is (for example, for Blackfin processors):

ccblkfn -proc ADSP-BF535 -mem main.c -o output.dxe

Invoking meminit with Callback Executables To directly invoke the memory initializer from a command line, use the -Init switch for each “callback” executable as shown below: meminit Input.dxe -o Output.dxe Init Callback1.dxe -Init Callback2.dxe

From the VisualDSP++ IDDE, choose Project -> Project Options and select the Link page. Use the Additional options field to process callback executable files.

For example, if you have two callback executable files (callback1.dxe and callback2.dxe) and you wish to pass them to the memory initializer,

7-12 VisualDSP++ 5.0 Linker and Utilities Manual Memory Initializer

enter them in the Additional options box as: -meminit -flag-meminit -Init callback1.dxe -Init callback2.dxe

Then click OK (see Figure 7-3).

Figure 7-3. Invoking Callback Executable from the VisualDSP++ IDDE

VisualDSP++ 5.0 Linker and Utilities Manual 7-13 Memory Initializer Command-Line Switches

Memory Initializer Command-Line Switches

Table 7-1 summarizes the memory initializer switches. It is followed by a detailed description of each switch. Most of the listed switches are optional. For a project in which the linker description file is well-defined (the .meminit and bsz_init memory sections are defined and the ZERO_INIT and RUNTIME_INIT qualifiers are set on the proper sections) and the sections are initialized properly in the source files, most of these optional switches may not be required. By default, the memory initializer automatically handles everything needed to create an initialization stream.

Table 7-1. Summary of Command-Line Options and Entries

Item Description

-BeginInit Initsymbol Specifies a symbol name for a variable that holds a pointer on page 7-15 pointing to the start address of an initialization stream.

-h[elp] on page 7-16 Displays the list of memory initializer switches.

-IgnoreSection Sectionname Directs the memory initializer to NOT process a section on page 7-16 selected in the primary input file.

-Init Initcode.dxe Specifies an executable file to be inserted into the initial- on page 7-16 ization stream and executed as a callback.

InputFile.dxe Specifies a primary input file. on page 7-17

-NoAuto Directs the memory initializer to NOT process sections in on page 7-17 the primary input file based on the section header flags. This switch is optional.

-NoErase Directs the memory initializer not to erase the data of the on page 7-17 processed sections in the primary executable file.

7-14 VisualDSP++ 5.0 Linker and Utilities Manual Memory Initializer

Table 7-1. Summary of Command-Line Options and Entries (Cont’d)

Item Description

-o Outputfile.dxe Specifies an output file. on page 7-18

-Section Sectionname Specifies a section from which the data will be extracted by on page 7-18 the memory initializer. This switch can be repeated to specify a number of the sections from the specified input primary file.

-v (Verbose) Outputs status information as the memory ini- on page 7-18 tializer processes files.

The following sections provide the detailed descriptions of the command- line switches.

-BeginInit Initsymbol

The -BeginInit Initsymbol switch is used to specify a symbol name for a variable that holds a pointer to the start address of an initialization stream. The memory initializer updates this pointer with the start address of the initialization stream produced by the memory initializer.

If this switch is absent, the default symbol name “___inits” (it has three leading underscores, when called from assembly code) is searched, which, by default, is in the bsz_init memory section. If this symbol cannot be found in the input primary file, an error message is issued; for example:

meminit -BeginInit boggy input.dxe ERROR: The specified destination section, .meminit, not found in the input file

If a symbol other than “___inits” is specified using this switch in a section other than “bsz_init”, the symbol must not be in any of the sections specified via the -Section Sectionname switch (on page 7-18). It also must be able to hold a value that is no less than the maximum address value for the particular processor. The run-time library provides a default

VisualDSP++ 5.0 Linker and Utilities Manual 7-15 Memory Initializer Command-Line Switches

symbol of “___inits” for the memory initializer and, therefore, it is not necessary to use this switch in most cases. This switch has no effect on callback executable files specified using the “-Init Initcode.dxe” on page 7-16.

-h[elp]

The -h[elp] switch displays the list of memory initializer switches.

-IgnoreSection Sectionname

The -IgnoreSection Sectionname switch is used to specify a section that is not to be processed by the memory initializer. This switch can be repeated to specify a number of sections not to be processed in the primary input file. All the specified sections must exist in the primary input file.

The -IgnoreSection switch is optional. It is normally easier to remove a section’s initialization qualifier (ZERO_INIT or RUNTIME_INIT) from the .ldf file than to use this switch. This switch does not affect a callback executable file specified using the -Init Sectionname switch.

-Init Initcode.dxe

The -Init Initcode.dxe switch is used to specify an executable file to be inserted into the initialization stream and executed as a callback. Any number of executable files can be specified this way, and it is allowed to specify the same file name a number of times. The callback executable file must exist before the memory initializer is run. All the code and data from callback executable files are extracted to make up the initialization stream. This is an optional switch.

7-16 VisualDSP++ 5.0 Linker and Utilities Manual Memory Initializer

InputFile.dxe

The InputFile.dxe parameter is used to specify a primary input file. The memory initializer issues an error message if no primary input file is specified.

-NoAuto

The -NoAuto switch directs the memory initializer to not process sections in the primary input file based on the section header flags (the section specified as either ZERO_INIT and RUNTIME_INIT qualifier in the .ldf file), but to only process sections specified on the command line using the -section SectionName switch. By default, the memory initializer automatically processes only the sections with ZERO_INIT and RUNTIME_INIT qualifiers in the .ldf file. This switch has no effect on the code and data of callback executable files specified using the -init switch. All the code and data sections of a callback executable file are processed by the memory initializer regardless whether this switch is used. This switch is optional.

-NoErase

The -NoErase switch directs the memory initializer not to erase the data of the processed sections. By default, the memory initializer empties the sections from which the data are extracted to create the initialization stream. This switch is valid for the primary input file only and has no effect on callback executable files. The memory initializer does not carry any sections of a callback executable file over to the output file, nor erase any sections, but only extracts the code and data from it to form the initialization stream.

VisualDSP++ 5.0 Linker and Utilities Manual 7-17 Memory Initializer Command-Line Switches

-o Outputfile.dxe

The -o Outputfile.dxe switch is used to specify an output file. If this switch is absent, the memory initializer makes an output file name from the root of the input file name. For example, if the input file name is InputFile.dxe, the output file name is created as InputFile1.dxe. This switch is optional.

-Section Sectionname

The -Section Sectionname switch is used to specify a section from which the data is extracted by the memory initializer. This switch can be repeated to specify a number of the sections from the specified input primary file. All the section specified must exist in the specified input primary file. Note that the section name specified via the -IgnoreSection switches cannot be used with the -Section switch. It is not necessary to use this switch to specify sections that already have the ZERO_INIT or RUNTIME_INIT qualifiers in the linker description file (.ldf), as the memory initializer processes such sections automatically. Using initialization qualifiers in the .ldf file is usually the simpler and recommended method. The -Section SectionName switch has no effect on callback executable files specified via the -Init switch. Therefore, do not use this switch to specify any sections in callback executable files.

-v

The -v or -verbose (verbose) switch directs the memory initializer to output status information as it processes files.

7-18 VisualDSP++ 5.0 Linker and Utilities Manual A FILE FORMATS

The VisualDSP++ development tools support many file formats. In some cases, several file formats for each development tool are supported. This appendix describes file formats that are prepared as input for the tools and points out the features of files produced by the tools. This appendix discusses three types of file formats: • “Source Files” on page A-2 • “Build Files” on page A-5 • “Debugger Files” on page A-9 Most of the development tools use industry-standard file formats. Sources that describe these formats appear in “Format References” on page A-10.

VisualDSP++ 5.0 Linker and Utilities Manual A-1 Source Files

Source Files

This section describes these input file formats: • “C/C++ Source Files” on page A-2 • “Assembly Source Files (.asm)” on page A-3 • “Assembly Initialization Data Files (.dat)” on page A-3 • “Header Files (.h)” on page A-4 • “Linker Description Files (.ldf)” on page A-4 • “Linker Command-Line Files (.txt)” on page A-5

C/C++ Source Files

C and C++ source files are text files (with extensions such as .c, .cpp, .cxx, and so on) that contain C/C++ code, compiler directives, possibly a mixture of assembly code and directives, and (typically) preprocessor commands. Several “dialects” of C code are supported: pure (portable) ANSI C, and at least two subtypes1 of ANSI C with Analog Devices extensions. These extensions include memory type designations for certain data objects, and segment directives used by the linker to structure and place executable files. For information on using the C/C++ compiler and associated tools, as well as a definition of Analog Devices extensions to ANSI C, refer to the VisualDSP++ C/C++ Compiler and Library Manual for appropriate target architecture.

1 With and without built-in function support; a minimal differentiator. There are others.

A-2 VisualDSP++ 5.0 Linker and Utilities Manual File Formats

Assembly Source Files (.asm)

Assembly source files are text files that contain assembly instructions, assembler directives, and (optionally) preprocessor commands. For information on assembly instructions, see your processor’s Programming Reference. The instruction set is supplemented with assembler directives. Preprocessor commands control macro processing and conditional assembly or compilation. For information on the assembler and preprocessor, refer to the VisualDSP++ Assembler and Preprocessor Manual.

Assembly Initialization Data Files (.dat)

Assembly initialization data (.dat) files are text files that contain fixed-point or floating-point data. These files provide the initialization data for an assembler .var directive or serve in other tool operations.

When a .var directive uses a .dat file for data initialization, the assembler reads the data file and initializes the buffer in the output object (.doj) file. Data files have one data value per line and may have any number of lines.

The .dat extension is explanatory or mnemonic. A directive to #include can take any file name (or extension) as an argument. Fixed-point values (integers) in data files may be signed, and they may be decimal-, hexadecimal-, octal-, or binary-base values. The assembler uses the prefix conventions in Table A-1 to distinguish between numeric formats. For all numeric bases, the assembler uses 16-bit words for data storage; 24-bit data is for the program code only. The largest word in the buffer determines the size for all words in the buffer. If there is some 8-bit data in a 16-bit-wide buffer, the assembler loads the equivalent 8-bit value into

VisualDSP++ 5.0 Linker and Utilities Manual A-3 Source Files

the most significant eight bits in the 8-bit memory location and zero-fills the lower eight bits.

Table A-1. Numeric Formats

Convention Description

0xnumber Hexadecimal number H#number h#number

number Decimal number D#number d#number

B#number Binary number. b#number

O#number Octal number. o#number

Header Files (.h)

Header files are ASCII text files that contain macros or other preprocessor commands that the preprocessor substitutes into source files. For information on macros or other preprocessor commands, refer to the VisualDSP++ C/C++ Compiler and Library Manual for appropriate target architecture. For information on the assembler and preprocessor, see the VisualDSP++ Assembler and Preprocessor Manual.

Linker Description Files (.ldf)

Linker description files are ASCII text files that contain commands for the linker in the linker’s scripting language. For information on this scripting language, see “LDF Commands” on page 3-36.

A-4 VisualDSP++ 5.0 Linker and Utilities Manual File Formats

Linker Command-Line Files (.txt)

Linker command-line files are ASCII text files that contain command-line input for the linker. For more information on the linker command line, see “Linker Command-Line Reference” on page 2-44.

Build Files

Build files are produced by the VisualDSP++ development tools when building a project. This section describes these build file formats: • “Assembler Object Files (.doj)” on page A-5 • “Library Files (.dlb)” on page A-6 • “Linker Output Files (.dxe, .sm, and .ovl)” on page A-6 • “Memory Map Files (.xml)” on page A-6 • “Loader Output Files in Intel Hex-32 Format (.ldr)” on page A-6 • “Splitter Output Files in ASCII Format (.ldr)” on page A-8

Assembler Object Files (.doj)

Assembler output object (.doj) files are in binary, executable and linkable file (ELF) format. Object files contain relocatable code and debugging information for a DSP program’s memory segments. The linker processes object files into an executable (.dxe) file. For information on the object file’s ELF format, see “Format References” on page A-10.

VisualDSP++ 5.0 Linker and Utilities Manual A-5 Build Files

Library Files (.dlb)

Library files, the archiver’s output, are in binary, executable and linkable file (ELF) format. Library files (called archive files in previous software releases) contain one or more object files (archive elements). The linker searches through library files for library members used by the code. For information on the ELF format used for executable files, refer to “Format References” on page A-10.

Linker Output Files (.dxe, .sm, and .ovl)

The linker’s output files are in binary, executable and linkable file (ELF) format. These executable files contain program code and debugging information. The linker fully resolves addresses in executable files. For information on the ELF format used for executable files, see the TIS Com- mittee texts cited in “Format References” on page A-10. The archiver automatically converts legacy input objects from L COFF to ELF format.

Memory Map Files (.xml)

The linker can output memory map files that contain memory and symbol information for your executable file(s). The memory map file contains a summary of memory defined with MEMORY{} commands in the .ldf file, and provides a list of the absolute addresses of all symbols. Memory map files are available only in .xml format.

Loader Output Files in Intel Hex-32 Format (.ldr)

The loader can output Intel hex-32 format (.ldr) files. These files support 8-bit-wide PROMs. The files are used with an industry-standard PROM programmer to program memory devices for a hardware system. One file contains data for the whole series of memory chips to be programmed.

A-6 VisualDSP++ 5.0 Linker and Utilities Manual File Formats

The following example shows how the Intel hex-32 format appears in the loader’s output file. Each line in the Intel hex-32 file contains an extended linear address record, a data record, or an end-of-file record.

:020000040000FA Extended linear address record

:0402100000FE03F0F9 Data record

:00000001FF End-of-file record Extended linear address records are used because data records have a 4-character (16-bit) address field, but in many cases, the required PROM size is greater than or equal to 0xFFFF bytes. Extended linear address records specify bits 16-31 for the data records that follow. Table A-2 shows an example of an extended linear address record.

Table A-2. Extended Linear Address Record Example

Field Purpose

:020000040000FA Example record

: Start character

02 Byte count (always 02)

0000 Address (always 0000)

04 Record type

0000 Offset address

FA Checksum

VisualDSP++ 5.0 Linker and Utilities Manual A-7 Build Files

Table A-3 shows the organization of an example data record, and Table A-4 shows an end-of-file record.

Table A-3. Data Record Example

Field Purpose

:0402100000FE03F0F9 Example record

: Start character

04 Byte count of this record

0210 Address

00 Record type

00 First data byte

F0 Last data byte

F9 Checksum

Table A-4. End-of-File Record Example

Field Purpose

:00000001FF End-of-file record

: Start character

00 Byte count (zero for this record)

0000 Address of first byte

01 Record type

FF Checksum

For more information, refer to the VisualDSP++ Loader and Utilities Manual.

Splitter Output Files in ASCII Format (.ldr)

When the loader is invoked as a splitter, its output can be an ASCII format file. ASCII format files are text representations of ROM memory

A-8 VisualDSP++ 5.0 Linker and Utilities Manual File Formats

images that you can use in post-processing. For more information, refer to no-boot mode information in the VisualDSP++ Loader and Utilities Manual.

Debugger Files

Debugger files provide input to the debugger to define simulation or emulation support of your program. The debugger supports all the executable file types produced by the linker (.dxe, .sm, .ovl). To simulate I/O, the debugger also supports the assembler’s data file (.dat) format and the loader’s loadable file (.ldr) formats. The standard hexadecimal format for a SPORT data file is one integer value per line. Hexadecimal numbers do not require a 0x prefix. A value can have any number of digits, but is read into the SPORT register as: • The hexadecimal number which is converted to binary • The number of binary bits read which matches the word size set for the SPORT register, which starts reading from the LSB. The SPORT register then fills with zero values shorter than the word size or conversely truncates bits beyond the word size on the MSB end. Example: In this example, a SPORT register is set for 20-bit words and the data file contains hexadecimal numbers. The simulator converts the HEX numbers

VisualDSP++ 5.0 Linker and Utilities Manual A-9 Format References

to binary and then fills or truncates to match the SPORT word size. In Table A-5, the A5A5 number is filled and 123456 is truncated.

Table A-5. SPORT Data File Example

Hex Number Binary Number Truncated/Filled

A5A5A 1010 0101 1010 0101 1010 1010 0101 1010 0101 1010

FFFF1 1111 1111 1111 1111 0001 1111 1111 1111 1111 0001

A5A5 1010 0101 1010 0101 0000 1010 0101 1010 0101

5A5A5 0101 1010 0101 1010 0101 0101 1010 0101 1010 0101

11111 0001 0001 0001 0001 0001 0001 0001 0001 0001 0001

123456 0001 0010 0011 0100 0101 0110 0010 0011 0100 0101 0110

Format References

The following texts define industry-standard file formats supported by VisualDSP++. • Gircys, G.R. (1988) Understanding and Using COFF by O’Reilly & Associates, Newton, MA • (1993) Executable and Linkable Format (ELF) V1.1 from the Portable Formats Specification V1.1, Tools Interface Standards (TIS) Committee

Go to: http://developer.intel.com/and search on the text string ELF. • (1993) Debugging Information Format (DWARF) V1.1 from the Portable Formats Specification V1.1, UNIX International, Inc.

Go to: http://developer.intel.com/ and search on the text string DWARF.

A-10 VisualDSP++ 5.0 Linker and Utilities Manual B UTILITIES

The VisualDSP++ development software includes the following utilities: • “elfdump – ELF File Dumper” • “elfpatch” • “plinker” elfdump – ELF File Dumper

The executable and linking format (ELF) file dumper (elfdump) utility extracts data from ELF-format executable (.dxe) files and yields text showing the ELF file’s contents.

The elfdump utility is often used with the archiver (elfar). Refer to “Disassembling a Library Member” on page B-3 for details. Also refer to “Dumping Overlay Library Files” on page B-4 on how to extract and view the contents of overlay library files. Syntax:

elfdump [switches] [objectfile]

VisualDSP++ 5.0 Linker and Utilities Manual B-1 elfdump – ELF File Dumper

Table B-1 shows switches used with the elfdump command.

Table B-1. ELF File Dumper Command-Line Switches

Switch Description

-fh Prints the file header

-arsym Prints the library symbol table

-arall Prints every library member

-help Prints the list of elfdump switches to stdout

-ph Prints the program header table

-sh Prints the section header table. This switch is the default when no options are specified.

-notes Prints note segment(s)

-n name Prints contents of the named section(s). The name may be a simple ‘glob’-style pattern, using “?” and “*” as wildcard characters. Each section’s name and type determines its output format, unless overridden by a modifier.

-i x0[-x1] Prints contents of sections numbered x0 through x1, where x0 and x1 are decimal integers, and x1 defaults to x0 if omitted. Formatting rules are the same as for the -n switch.

-all Prints everything. This is the same as -fh -ph -sh -notes -n ‘*’.

-ost Omits string table sections

-c Same as -ost (deprecated)

-s Same as -ost (deprecated)

-v Prints version information

objectfile Specifies the file whose contents are to be printed. It can be a core file, executable, shared library, or relocatable object file. If the name is in the form A(B), A is assumed to be a library and B is an ELF member of the library. B can be a pattern similar to the one accepted by -n.

B-2 VisualDSP++ 5.0 Linker and Utilities Manual Utilities

The -n and -i switches can have a modifier letter options after the main option character to force section contents to be formatted as:

• a – Dumps contents in hex and ASCII, 16 bytes per line.

• x – Dumps contents in hex, 32 bytes per line.

• xN – Dumps contents in hex, N bytes per group (default is N = 4).

• t – Dumps contents in hex, N bytes per line, where N is the section’s table entry size. If N is not in the range 1 to 32, 32 is used.

• hN – Dumps contents in hex, N bytes per group.

• HN – Dumps contents in hex, (MSB first order), N bytes per group.

• i – Prints contents as list of disassembled machine instructions.

• s – Prints contents as list of disassembled machine instructions and also prints labels.

Disassembling a Library Member

The elfar and elfdump utilities are more effective when their capabilities are combined. One application of these utilities is for disassembling a library member and converting it to source code. Use this technique when the source of a particularly useful routine is missing and is available only as a library routine.

For information about elfar, refer to “Archiver” on page 6-1. The following procedure lists the objects in a library, extracts an object, and converts the object to a listing file. The first archiver command line lists the objects in the library and writes the output to a text file. elfar -p libc.dlb > libc.txt

Open the text file, scroll through it, and locate the object file you need.

VisualDSP++ 5.0 Linker and Utilities Manual B-3 elfdump – ELF File Dumper

To convert the object file to an assembly listing file with labels, use the following elfdump command line, which references the library and the object file in the library. elfdump -ns * libc.dlb (fir.doj) > fir.asm

The output file is practically source code. Just remove the line numbers and opcodes. Disassembly yields a listing file with symbols. Assembly source with symbols can be useful if you are familiar with the code and hopefully have some documentation on what the code does. If the symbols are stripped during linking, the dumped file contains no symbols. Disassembling a third-party’s library may violate the license for the [ third-party software. Ensure there are no copyright or license issues with the code’s owner before using this disassembly technique.

Dumping Overlay Library Files

Use the elfar and elfdump utilities to extract and view the contents of overlay library (.ovl) files. For example, the following command lists (prints) the contents (library members) of the clone2.ovl library file. elfar -p clone2.ovl

The following command allows you to view one of the library members (clone2.elf). elfdump -all clone2.ovl(clone2.elf)

The following commands extract clone2.elf and print its contents.

elfar -e clone2.ovl clone2.elf elfdump -all clone2.elf L Switches for the elfdump commands are case sensitive.

B-4 VisualDSP++ 5.0 Linker and Utilities Manual Utilities elfpatch

The ELF patch (elfpatch) utility allows the bits of an ELF section to be extracted or replaced from a file. Syntax:

elfpatch -get [section-name] -o[output-bits-filename] -text [input-elf-filename]

elfpatch -replace [section-name] -o[output-filename] -bits [input-bits-filename] -text [input-elf-filename]

elfpatch [help | version]

Examples: elfpatch -get _ov_os_overlay_1 -o bytes_bin o1.ovl (overlay1.elf) elfpatch -get L1_code -o bytes_txt -text p0.dxe elfpatch -replace _ov_os_overlay_1 -o o1_new_from_txt.ovl -bits bytes_txt -text o1.ovl (overlay1.elf) elfpatch -replace L1_code -o p0_new.dxe -bits bytes_bin p0.dxe

Extracting a Section in an ELF File

The elfpatch -get command dumps the raw contents of a section without any additional formatting. The input-elf-filename parameter may be one of the following: • A stand-alone (non-archive) ELF file containing a section specified by the section-name parameter • A library (filename) combination

The -text switch specifies that the output should be a stream of printable text. Specifically, the output must be hexadecimal digital (with each one byte of binary output resulting in two bytes of text (hex) output. If the -o switch is not specified, the output (in bits) is written to stdout.

VisualDSP++ 5.0 Linker and Utilities Manual B-5 plinker

Replacing Raw Contents of a Section in an ELF File

The elfpatch -replace command replaces the raw contents of a section. The replacement bits need not be the same size as the section being replaced. If the replacement resulted in the replace section clobbering a por- L tion of another section, an error would result in a resolved ELF file.

If the -bits switch is not specified, bits are read from stdin.

The input-elf-filename parameter must exist and be either of the following: • A stand-alone (non-archive) ELF file containing a section specified by the section-name parameter • A library (filename) combination

Ultimately, the input-elf-filename parameter must contain a section specified by the section-name parameter. If the -o switch is not specified, the output (ELF file) is written to stdout.

The -text switch specifies that the input should be a stream of printable text. Specifically, the output must be hexadecimal digital (with each one byte of binary output resulting in two bytes of text (hex) output.

Standard input (stdin) and standard output (stdout) are used to L facilitate piping. Here is an example command line: elfpatch -get code input.dxe | my-transformation | elfpatch -replace code input.dxe -o output.dxe plinker

In VisualDSP++, the plinker command-line tool provides “partial linker” functionality.

B-6 VisualDSP++ 5.0 Linker and Utilities Manual Utilities

The plinker tool is a specialized linker that produces a partially-linked relocatable object file instead of a fully-linked executable. It is similar in function to a standard UNIX linker (ld) invoked with the -r switch. The partial linker performs two main functions: • It combines a number of input object files into a single output file by concatenating sections having the same name into a single section in the output file. All references to offsets or indices within the input files are modified to reflect the new locations of sections, and of records within sections, in the output file. • For symbols with external linkage, it resolves multiple occurrences of the same symbol to a single instance. Local function and object symbols are made unique by adding a suffix in order to satisfy linker input requirements. The partial linker can link object files and archive libraries. The ordering of object file and library arguments in the plinker command line is highly significant. If a referenced external symbol has no global definition and multiple weak definitions, the symbol is resolved to the first weak definition that is encountered.

For example, consider the following plinker command line:

plinker -o out.o in1.o in2.o -l lib1.a in3.o -l lib2.a

Library “lib1.a” will be searched only for references encountered in “in1.o;” and “in2.o”. It will not be searched again after “in3.o” and library “lib2.a” have been read. If either “in3.o” or “lib2.a” has references that should be resolved by “lib1.a”, the “lib1.a” must be specified a second time, later in the command line.

If invoked with no arguments, or with “-h” or “-help”, the partial linker prints command-line information. The -info switch provides additional details, including current anomalies and limitations. Tracing of linker

VisualDSP++ 5.0 Linker and Utilities Manual B-7 plinker

actions can be enabled and finely controlled with the -t, -tr, and -ntr switches.

B-8 VisualDSP++ 5.0 Linker and Utilities Manual C LDF PROGRAMMING EXAMPLES FOR BLACKFIN PROCESSORS

This appendix provides several typical .ldf files. used with Blackfin processors. As you modify these examples, refer to the syntax descriptions in “LDF Commands” on page 3-36. This appendix provides the following examples. • “Linking for a Single-Processor System” on page C-2 • “Linking Large Uninitialized or Zero-initialized Variables” on page C-4 The source code for several programs is bundled with the L development software. Each program includes an .ldf file. For working examples of the linking process, examine the .ldf files that come with the examples. These examples are in the directory: /Blackfin/examples.

The development software includes a variety of default .ldf files. L These files provide an example .ldf file for each processor’s internal memory architecture. The default .ldf files are in the directory: /Blackfin/ldf.

VisualDSP++ 5.0 Linker and Utilities Manual C-1 Linking for a Single-Processor System

Linking for a Single-Processor System

When you link an executable file for a single-processor system, the .ldf file describes the processor’s memory and places code for that processor. The .ldf file in Listing C-1 is for a single-processor system. Note the following commands in this example file.

• ARCHITECTURE() defines the processor type

• SEARCH_DIR() commands add the lib and current working directory to the search path

• $OBJS and $LIBS macros retrieve object (.doj) and library (.dlb) file input

• MAP() outputs a map file

• MEMORY{} defines memory for the processor

• PROCESSOR{} and SECTIONS{} commands define a processor and place program sections for that processor’s output file by using the memory definitions

Listing C-1. Example LDF for a Single-Processor System

ARCHITECTURE(ADSP-BF535)

SEARCH_DIR( $ADI_DSP/Blackfin/lib )

MAP(SINGLE-PROCESSOR.MAP) // Generate a MAP file

// $ADI_DSP is a predefined linker macro that expands // to the VDSP install directory. Search for objects in // directory Blackfin/lib relative to the install directory

C-2 VisualDSP++ 5.0 Linker and Utilities Manual LDF Programming Examples for Blackfin Processors

LIBS libc.dlb, libevent.dlb, libsftflt.dlb, libcpp_blkfn.dlb, libcpprt_blkfn.dlb, libdsp.dlb $LIBRARIES = LIBS, librt.dlb;

// single.doj is a user generated file. The linker will be // invoked as follows // linker -T single-processor.ldf single.doj. // $COMMAND_LINE_OBJECTS is a predefined linker macro // The linker expands this macro into the name(s) of the // the object(s) (.doj files) and archives (.dlb files) // that appear on the command line. In this example, // $COMMAND_LINE_OBJECTS = single.doj

$OBJECTS = $COMMAND_LINE_OBJECTS;

// A linker project to generate a DXE file

PROCESSOR P0 { OUTPUT( SINGLE.dxe ) // The name of the output file

MEMORY // Processor specific memory command { INCLUDE( "BF535_memory.ldf") }

SECTIONS // Specify the Output Sections { INCLUDE( "BF535_sections.ldf" } // end P0 sections } // end P0 processor

VisualDSP++ 5.0 Linker and Utilities Manual C-3 Linking Large Uninitialized or Zero-initialized Variables

Linking Large Uninitialized or Zero-initialized Variables

When linking an executable file that contains large uninitialized variables, use the NO_INIT (equivalent to SHT_NOBITS legacy qualifier) or ZERO_INIT section qualifier to reduce the file size. A variable defined in a source file normally takes up space in an object and executable file even if that variable is not explicitly initialized when defined. For large buffers, this action can result in large executables filled mostly with zeros. Such files take up excess disk space and can incur long download times when used with an emulator. This situation also may occur when you boot from a loader file (because of the increased file size). Listing C-2 shows an example of assembly source code. Listing C-3 shows the use of the NO_INIT and ZERO_INIT sections to avoid initialization of a segment.

The .ldf file can omit an output section from the output file. The NO_INIT qualifier directs the linker to omit data for that section from the output file. Refer to “SECTIONS{}” on page 3-61 for more information on L the NO_INIT and ZERO_INIT section qualifiers.

The NO_INIT qualifier corresponds to the /UNINIT segment qualifier L in previous (.ach) development tools. Even if you do not use NO_INIT, the boot loader removes variables initialized to zeros from the .ldr file and replaces them with instructions for the loader kernel to zero out the variable. This action reduces the loader’s output file size, but still requires execution time for the processor to initialize the memory with zeros.

Listing C-2. Large Uninitialized Variables: Assembly Source

.SECTION/NO_INIT extram_area; /* 1Mx8 EXTRAM */

C-4 VisualDSP++ 5.0 Linker and Utilities Manual LDF Programming Examples for Blackfin Processors

.BYTE huge_buffer[0x006000]; .SECTION/ZERO_INIT zero_extram_area; .BYTE huge_zero_buffer[0x006000];

Listing C-3. Large Uninitialized Variables: LDF Source

ARCHITECTURE(ADSP-BF535) $OBJECTS = $COMMAND_LINE_OBJECTS; // Libraries & objects from // the command line MEMORY { mem_extram { TYPE(RAM) START(0x10000) END(0x15fff) WIDTH(8) } // end segment } // end memory

PROCESSOR P0 { LINK_AGAINST( $COMMAND_LINE_LINK_AGAINST ) OUTPUT( $COMMAND_LINE_OUTPUT_FILE ) // NO_INIT section isn’t written to the output file SECTIONS { extram_output NO_INIT { INPUT_SECTIONS( $OBJECTS ( extram_area ) ) } >mem_extram zero_extram_output ZERO_INIT { INPUT_SECTIONS ( $OBJECTS ( zero_extram_area ) ) } >mem_extram } // end section } // end processor P0

VisualDSP++ 5.0 Linker and Utilities Manual C-5 Linking Large Uninitialized or Zero-initialized Variables

C-6 VisualDSP++ 5.0 Linker and Utilities Manual D LDF PROGRAMMING EXAMPLES FOR SHARC PROCESSORS

This appendix provides several typical .ldf files used with SHARC processors. As you modify these examples, refer to the syntax descriptions in “LDF Commands” on page 3-36. This appendix provides the following examples: • “Linking a Single-Processor SHARC System” on page D-2 • “Linking Large Uninitialized Variables” on page D-4 • “Linking for MP and Shared Memory” on page D-6 The source code for several programs is bundled with your devel- L opment software. Each program includes an .ldf file. For working examples of the linking process, examine the .ldf file that come with the examples. Examples are in the following directory. /21k/Examples.

A variety of processor-specific default .ldf files come with the L development software, providing information about each processor’s internal memory architecture. Default .ldf files are located in the following directory. /21k/ldf.

VisualDSP++ 5.0 Linker and Utilities Manual D-1 Linking a Single-Processor SHARC System

Linking a Single-Processor SHARC System

When linking an executable for a single-processor system, the .ldf file describes the processor’s memory and places code for that processor. Listing D-1 shows a single-processor .ldf file. Note the following commands in this file:

• ARCHITECTURE() defines the processor type.

• SEARCH_DIR() adds the lib and current working directory to the search path.

• $OBJS and $LIBS macros get object (.doj) and library (.dlb) file input.

• MAP() outputs a map file.

• MEMORY{} defines memory for the processor.

• PROCESSOR{} and SECTIONS{} defines a processor and place program sections for that processor’s output file, using the memory definitions.

Listing D-1. Single-Processor System LDF Example

// Link for the ADSP-21161 ARCHITECTURE(ADSP-21161) SEARCH_DIR ( $ADI_DSP/211xx/lib ) MAP (SINGLE-PROCESSOR.XML) // Generate a MAP file

// $ADI_DSP is a predefined linker macro that expands to // the VisualDSP++ installation directory. Search for objects // in directory 21k/lib relative to the installation directory

D-2 VisualDSP++ 5.0 Linker and Utilities Manual LDF Programming Examples for SHARC Processors

// lib161.dlb is an ADSP-2116x-specific library // and must precede libc.dlb, the C library // to link the 2116x-specific routines.

$LIBS = lib161.dlb, libc.dlb;

// 161_hdr.doj is the standard initialization file for 2116x $OBJS = $COMMAND_LINE_OBJECTS, 161_hdr.doj;

// A linker project to generate a .dxe file PROCESSOR P0 { OUTPUT ( ./SINGLE.dxe ) // The name of the output file

MEMORY // Processor-specific memory command { INCLUDE("21161_memory.h") }

SECTIONS // Specify the output sections { INCLUDE( "21161_sections.h" ) } // end P0 sections } // end P0 processor

VisualDSP++ 5.0 Linker and Utilities Manual D-3 Linking Large Uninitialized Variables

Linking Large Uninitialized Variables

When linking an executable file that contains large uninitialized variables, use the NO_INIT (equivalent to SHT_NOBITS legacy qualifier) or ZERO_INIT section qualifier to reduce the file size. A variable defined in a source file normally takes up space in an object and executable file even if that variable is not explicitly initialized when defined. For large buffers, this action can result in large executables filled mostly with zeros. Such files take up excess disk space and can incur long download times when used with an emulator. This situation also may occur when you boot from a loader file (because of the increased file size). Listing D-2 shows an example of assembly source code. Listing D-3 shows the use of the NO_INIT and ZERO_INIT sections to avoid initialization of a segment.

The NO_INIT qualifier corresponds to the /UNINIT segment qualifier L in previous (.ach) development tools. Even if NO_INIT is not used, the boot loader removes variables initialized to zeros from the .ldr file and replaces them with instructions for the loader kernel to zero-out the variable. This action reduces the loader’s output file size, but still requires execution time for the processor to initialize the memory with zeros.

Listing D-2. Large Uninitialized Variables: Assembly Source

.SECTION/DM/NO_INIT sdram_area; /* 1Mx32 SDRAM */ .VAR huge_buffer[0x100000];

D-4 VisualDSP++ 5.0 Linker and Utilities Manual LDF Programming Examples for SHARC Processors

Listing D-3. Large Uninitialized Variables: LDF Source

ARCHITECTURE(ADSP-21161) $OBJECTS = $COMMAND_LINE_OBJECTS; // Libraries & objects from // the command line MEMORY { mem_sdram { TYPE(DM RAM) START(0x3000000) END(0x30FFFFF) WIDTH(32) } // end segment } // end memory

PROCESSOR P0 { LINK_AGAINST( $COMMAND_LINE_LINK_AGAINST ) OUTPUT( $COMMAND_LINE_OUTPUT_FILE ) // NO_INIT section isn’t written to the output file SECTIONS { sdram_output NO_INIT { INPUT_SECTIONS( $OBJECTS ( sdram_area ) ) } >mem_sdram zero_sdram_output ZERO_INIT { INPUT_SECTIONS ( $OBJECTS ( zero_sdram_area ) ) } >mem_sdram } // end section } // end processor P0

VisualDSP++ 5.0 Linker and Utilities Manual D-5 Linking for MP and Shared Memory

Linking for MP and Shared Memory

When linking executable files for a multiprocessor system using shared memory, the .ldf file describes the multiprocessor memory offsets, shared memory, each processor’s memory, and places code for each processor. Here are the major commands in an .ldf file. For examples, examine the .ldf files included with the installation CD.:

• The ARCHITECTURE() command defines the processor type, which can be one type only.

• The SEARCH_DIR() command adds the lib and current working directory to the search path.

• The $OBJS and $LIBS macros get object (.doj) and library (.dlb) file input.

• The MPMEMORY{} command defines each processor’s offset within multiprocessor memory.

• The SHARED_MEMORY{} command identifies the output for the shared memory items.

• The MAP() command outputs map files.

• The MEMORY{} command defines memory for the processors.

• The PROCESSOR{} and SECTIONS{} commands define each processor and place program sections using memory definitions for each processor’s output file.

• The LINK_AGAINST() commands resolve symbols within multiprocessor memory.

D-6 VisualDSP++ 5.0 Linker and Utilities Manual LDF Programming Examples for SHARC Processors

Listing D-4. LDF for Multiprocessor System With Shared Memory

ARCHITECTURE(ADSP-21062)

// // ADSP-21062 Memory Map: // ------// Internal memory 0x0000 0000 to 0x0007 ffff // ------// 0x0000 0000 to 0x0000 00ff IOP Regs // 0x0000 0100 to 0x0001 ffff (reserved) // Block 0 0x0002 0000 to 0x0002 7fff // Normal Word (32/48) Addresses // (0x0002 0000 to 0x0002 4fff) // 48-bit words // (0x0002 0000 to 0x0002 7fff) // 32-bit words // Block 1 0x0002 8000 to 0x0002 ffff // Normal Word (32/48) Addresses // (0x0002 8000 to 0x0002 cfff) // 48-bit words // (0x0002 8000 to 0x0002 ffff) // 32-bit words // alias of Block 1 0x0003 0000 to 0x0003 7fff // Normal Word (32/48) Addresses // alias of Block 1 0x0003 8000 to 0x0003 ffff // Normal Word (32/48) Addresses // Block 0 0x0004 0000 to 0x0004 ffff // Short Word (16) Addresses // Block 1 0x0005 0000 to 0x0005 ffff // Short Word (16) Addresses // alias of Block 1 0x0006 0000 to 0x0006 ffff // Short Word (16) Addresses

VisualDSP++ 5.0 Linker and Utilities Manual D-7 Linking for MP and Shared Memory

// alias of Block 1 0x0007 0000 to 0x0007 ffff // Short Word (16) Addresses // ------// Multiproc memory 0x0008 0000 to 0x003f ffff // ------// 0x0008 0000 to 0x000f ffff SHARC ID=001 Internal memory // 0x0010 0000 to 0x0017 ffff SHARC ID=010 Internal memory // 0x0018 0000 to 0x001f ffff SHARC ID=011 Internal memory // 0x0020 0000 to 0x0027 ffff SHARC ID=100 Internal memory // 0x0028 0000 to 0x002f ffff SHARC ID=101 Internal memory // 0x0030 0000 to 0x0037 ffff SHARC ID=110 Internal memory // 0x0038 0000 to 0x003f ffff SHARC ID=all Internal memory // ------// External memory 0x0040 0000 to 0xffff ffff // ------// // This architecture file allocates: // Internal // External

SEARCH_DIR( $ADI_DSP\21k\lib)

$LIBRARIES = lib060.dlb ;

$OBJECTS = $COMMAND_LINE_OBJECTS ;

//MAP(ffton2pe.map)

// Memory architecture description for FFT example on a 21062. // 256 48-bit words for interrupt vector table // (reset vector location). // 5888 48-bit words of program memory for code storage. // 2048 48-bit words of second segment of program memory // for code storage.

D-8 VisualDSP++ 5.0 Linker and Utilities Manual LDF Programming Examples for SHARC Processors

// 20k 32-bit words of internal program memory // for data storage. // 32k 32-bit words of internal data memory for data storage. // 4k 32-bit words of first segment of external data memory // for data storage. // 4k 32-bit words of second segment of external data memory // for data storage.

MEMORY { isr_tabl { TYPE(PM RAM) START(0x00020000) END(0x000200FF) WIDTH(48) } pm_code { TYPE(PM RAM) START(0x00020100) END(0x00021fff) WIDTH(48) } pm_data { TYPE(PM RAM) START(0x00023000) END(0x00027fff) WIDTH(32) }

dm_data { TYPE(DM RAM) START(0x00028000) END(0x0002ffff) WIDTH(32) }

Ext_idat { TYPE(PM RAM) START (0x400000) END (0x400FFF) WIDTH(32) } Ext_rdat { TYPE(DM RAM) START (0x401000) END (0x401FFE) WIDTH(32) }

}// End MEMORY

// Declare offset for multiprocessor memory space // of the 2 processors. MPMEMORY { p1 { START(0x00080000) } p2 { START(0x00100000) }

VisualDSP++ 5.0 Linker and Utilities Manual D-9 Linking for MP and Shared Memory

}

// Generate an object file which contains memory to be shared // by mutliple processors. // SHARED_MEMORY { OUTPUT($COMMAND_LINE_OUTPUT_DIRECTORY\common.sm) // SHARED MEMORY output file name.

// Map the sections specified in the program files // to sections declared in SHARED MEMORY and // use these sections to create the *.sm file. SECTIONS { .Ext_idat { INPUT_SECTIONS( shared_mem.doj(Ext_idat) ) } Ext_idat

.Ext_rdat { INPUT_SECTIONS( shared_mem.doj(Ext_rdat) ) } Ext_rdat

} //End Sections } // End Shared Memory

PROCESSOR p1 { LINK_AGAINST( $COMMAND_LINE_OUTPUT_DIRECTORY\common.sm , $COMMAND_LINE_OUTPUT_DIRECTORY\Fftrad2m.dxe ) // Other DXE and SM files to link against. OUTPUT( $COMMAND_LINE_OUTPUT_DIRECTORY\Fftrad2s.dxe ) // Shared memory and P2 executable.

D-10 VisualDSP++ 5.0 Linker and Utilities Manual LDF Programming Examples for SHARC Processors

SECTIONS { INCLUDE( "pe1_sections.h") // The file containing SECTIONs definition }// End SECTIONs

}// End p1

PROCESSOR p2 { LINK_AGAINST($COMMAND_LINE_OUTPUT_DIRECTORY\common.sm , $COMMAND_LINE_OUTPUT_DIRECTORY\Fftrad2s.dxe ) // Other DXE and SM files to link against, OUTPUT( $COMMAND_LINE_OUTPUT_DIRECTORY\Fftrad2m.dxe ) // Shared memory and P1 executable. SECTIONS { INCLUDE( "pe2_sections.h") // The file containing SECTIONs definition }// End Sections }// End p2

VisualDSP++ 5.0 Linker and Utilities Manual D-11 Linking for MP and Shared Memory

Reflective Semaphores

Semaphores may be used in multiprocessor (MP) systems to permit processors to share resources such as memory or I/O. A semaphore is a flag that can be read and written by any of the processors sharing the resource. A semaphore’s value indicates when the processor can access the resource. Reflective semaphores permit communication among processors that share a multiprocessor memory space. Use broadcast writes to implement reflective semaphores in an MP system. Broadcast writes allow simultaneous transmission of data to all the SHARC processors in an MP system. The master processor can broadcast writes to the same memory location or IOP register on all the slaves. During a broadcast write, the master also writes to itself unless the broadcast is a DMA write. Broadcast writes can also be used to simultaneously download code or data to multiple processors. Bus lock can be used in combination with broadcast writes to implement reflective semaphores in an MP system. The reflective semaphore should be located at the same address in internal memory (or IOP register) of each SHARC processor.

SHARC processors have a “broadcast” space. Use .ldf files (or header files) to define a memory segment in this space, just as in internal memory or any processor MP space. The broadcast space aliases internal space, so if there is a memory segment defined in the broadcast space, the .ldf file cannot have a memory segment at the corresponding address in the internal space (or in the MP space of any processor). Otherwise, the linker generates an error indicating that the memory definition is not valid. To check the semaphore, each SHARC processor reads from its own internal memory. Any object in the project can be mapped to an appropriate memory segment defined in the broadcast space for use as a reflective

D-12 VisualDSP++ 5.0 Linker and Utilities Manual LDF Programming Examples for SHARC Processors

semaphore. If an object defining symbol SemA is mapped to a broadcast space, when the program writes to SemA, the written value appears at the aliased internal address of each processor in the cluster. Each processor may read the value using SemA, or read it from internal memory by selecting (SemA-0x380000), thus avoiding bus traffic. To modify the semaphore, a SHARC processor requests bus lock and then performs a broadcast write to the semaphore address (for example, SemA). The processors should read the semaphore before modifying it to L verify that another processor has not changed it. For more information on semaphores, refer to your processor’s Hardware Reference manual.

VisualDSP++ 5.0 Linker and Utilities Manual D-13 Linking for MP and Shared Memory

D-14 VisualDSP++ 5.0 Linker and Utilities Manual E LDF PROGRAMMING EXAMPLES FOR TIGERSHARC PROCESSORS

This appendix provides several typical .ldf file used with TigerSHARC processors. As you modify these examples, refer to the syntax descriptions in “LDF Commands” on page 3-36. This appendix provides information about the following: • “Linking a Single-Processor System” on page E-2 • “Linking Large Uninitialized or Zero-Initialized Variables” on page E-4 • “Linking an ADSP-TS101 MP Shared Memory System” on page E-6 The source code for several programs is bundled with your devel- L opment software. Each program includes an .ldf file. For working examples of the linking process, examine the .ldf files that come with the examples. Examples are in the following directory. /TS/Examples.

A variety of processor-specific default .ldf file come with the L development software, providing information about each processor’s internal memory architecture. Default .ldf files are located in the following directory. /TS/ldf.

VisualDSP++ 5.0 Linker and Utilities Manual E-1 Linking a Single-Processor System

Linking a Single-Processor System

When linking an executable for a single-processor system, the .ldf file describes the processor’s memory and places code for that processor. The .ldf file in Listing E-1 shows a single-processor .ldf file. Note the following commands in this file:

• ARCHITECTURE() defines the processor type.

• SEARCH_DIR() adds the lib and current working directory to the search path.

• $OBJS and $LIBS macros get object (.doj) and library (.dlb) file input.

• MAP() outputs a map file.

• MEMORY{} defines memory for the processor.

• PROCESSOR{} and SECTIONS{} defines a processor and place program sections for that processor’s output file, using the memory definitions.

Listing E-1. Single-Processor System LDF Example

ARCHITECTURE(ADSP-TS201)

SEARCH_DIR ( $ADI_DSP/TS/lib )

MAP (SINGLE-PROCESSOR.xml) // Generate a MAP file

// $ADI_DSP is a predefined linker macro that expands to // the VisualDSP++ installation directory. Search for objects // in directory TS/lib relative to the installation directory

E-2 VisualDSP++ 5.0 Linker and Utilities Manual LDF Programming Examples for TigerSHARC Processors

$LIBS = libc.dlb;

// ts_header.doj is the standard initialization file for TSxxx

$OBJS = $COMMAND_LINE_OBJECTS, ts_hdr.doj;

// A linker project to generate a .dxe file

PROCESSOR P0 { OUTPUT ( ./SINGLE.dxe ) // The name of the output file

MEMORY // Processor-specific memory command { INCLUDE("TS201_memory.ldf") }

SECTIONS // Specify the output sections { INCLUDE( "TS201_sections.ldf" ) } // end P0 sections } // end P0 processor

VisualDSP++ 5.0 Linker and Utilities Manual E-3 Linking Large Uninitialized or Zero-Initialized Variables

Linking Large Uninitialized or Zero-Initialized Variables

When linking an executable file that contains large uninitialized variables, use the NO_INIT (equivalent to SHT_NOBITS legacy qualifier) or ZERO_INIT section qualifier to reduce the file size. A variable defined in a source file normally takes up space in an object and executable file even if that variable is not explicitly initialized when defined. For large buffers, this action can result in large executables filled mostly with zeros. Such files take up excess disk space and can incur long download times when used with an emulator. This situation also may occur when you boot from a loader file (because of the increased file size). Listing E-2 shows an example of assembly source code. Listing E-3 shows the use of the NO_INIT and ZERO_INIT sections to avoid initialization of a segment.

E-4 VisualDSP++ 5.0 Linker and Utilities Manual LDF Programming Examples for TigerSHARC Processors

Listing E-2. Large Uninitialized Variables: Assembly Source

.SECTION/NO_INIT sdram_area; /* 1Mx32 SDRAM */ .VAR huge_buffer[0x1000000];

Listing E-3. Large Uninitialized Variables: LDF Source

ARCHITECTURE(ADSP-TS201) $OBJECTS = $COMMAND_LINE_OBJECTS; // Libraries & objects from // the command line MEMORY { SDRAM { TYPE(RAM) START(0x04000000) END(0x07FFFFFF) WIDTH(32) } // end segment } // end memory

VisualDSP++ 5.0 Linker and Utilities Manual E-5 Linking an ADSP-TS101 MP Shared Memory System

Linking an ADSP-TS101 MP Shared Memory System

• The ARCHITECTURE() command defines the processor type, which can only be one type.

• The MPMEMORY{} command defines each processor’s offset within multiprocessor memory.

• The SHARED_MEMORY{} command identifies the output for the shared memory items.

• The MEMORY{} command defines memory for the processors.

• The PROCESSOR{} and SECTIONS{} commands define each processor and place program sections using memory definitions for each processor’s output file.

• The LINK_AGAINST() commands resolve symbols within multiprocessor memory.

E-6 VisualDSP++ 5.0 Linker and Utilities Manual IINDEX

A archiver -a archiver switch, 6-15 about, 6-1 absolute data placements, 2-59 accessing archived functions, 6-6 ABSOLUTE() LDF operator, 3-23 command-line switches and parameters, 6-15 address space, allocating, 3-56 command-line syntax, 6-14, 6-17 ADDR() LDF operator, 3-24 handling arbitrary files, 6-2 $ADI_DSP LDF macro, 3-32 in code disassembly, B-3 ADSP-21xxx processors, See SHARC processors symbol encryption, 6-11, 6-12, 6-15 ADSP-TSxxx processors, See TigerSHARC tagging with version, 6-7, 6-8 processors version information, 6-7, 6-9, 6-10 ALGORITHM() LDF command, 3-71 wildcard character, 6-6, 6-16 ALIGN() LDF command, 3-37 arguments, passing for simulation or emulation, alignment properties, managing, 4-63 2-44 ALL_FIT LDF identifier, 3-71, 4-66 _argv_string null-terminated string, 2-44 -anv archiver switch, 6-15 .asm files, assembler source, A-3 ARCHITECTURE() LDF command, 3-38 assembler archive directives with archiver, 6-5 files, See also library files initialization data files (.dat), A-3 members, A-6 object files (.doj), A-5 routines, 6-4 source files (.asm), 1-3, A-3 specify objects in, 6-6 attributes, used for linking, 2-42, 2-43 viewing files, B-4 writing library files, 6-4 B -BeginInit InitSymbol switch, 7-15 BEST_FIT LDF identifier, 3-72

VisualDSP++ 5.0 Linker and Utilites Manual I-1

Index

Blackfin memory sections Blackfin processors bsz_init input section, 2-23 basic .ldf file example, 3-7 .cht input section, 2-25 byte addressing, 5-10 constdata input section, 2-22 customized .ldf file, 3-3 cplb_code input section, 2-22 .ldf file programming examples, C-1 cplb_data input section, 2-22 memory configurations, 2-32, 3-9 cplb input section, 2-25 branch ctor input section, 2-22 expansion instruction, 2-59, 2-60, 2-63 ctorl input section, 2-24 instructions, 5-36 data1 input section, 2-22 breakpoints, on overlays, 5-7 .edt input section, 2-25 broadcast exception handling data sections, 2-24 space, D-12 .frt input section, 2-25 writes, D-12 .gdt, .gdtl, .frt, .cht, .edt, and .rtti, 2-24 bsz_init memory section, serving memory .gdtl input section, 2-25 initializer, 7-8 heap input section, 2-23 buffers, allocating to different memory L1_bcz input section, 2-26 segments, 2-42 L1_code input section, 2-26 build errors, linker, 2-10 L1_DATA_A input section, 2-22, 2-26 build files, A-5 L1_DATA_B input section, 2-22, 2-26 built-in LDF macros, 3-30 L1_data input section, 2-26 bus lock L2_bcz input section, 2-26 broadcast writes, D-12 l2_shared input section, 2-27 multiprocessor systems, D-12 L2_sram_a input section, 2-27 byte L2_sram_b input section, 2-27 order, 4-61 L2_sram input section, 2-26 packing, 4-62 l2_sram input section, 2-26 mc_data input section, 2-25 C noncache_code input section, 2-23 primio_atomic_lock input section, 2-25 C++ program input section, 2-21 constructor functions, 2-22 .rtti input section, 2-25 virtual function tables, 2-17, 2-21, 2-24 sdram0_bank input section, 2-24 caching, external memory, 3-9 sdram0 input section, 2-23 callback executable file, 7-3, 7-5, 7-16 sdram_bcz input section, 2-24 calls sdram_shared input section, 2-24 inter-overlay, 5-27 stack input section, 2-23 inter-processor, 5-28 voldata input section, 2-22 -c archiver switch, 6-15 vtbl input section, 2-24 C/C++ source files, A-2 character identifier (.), 1-8 .cht input section, 2-14

I-2 VisualDSP++ 5.0 Linker and Utilites Manual

Index

color selection, in Expert Linker, 4-13 data1-2 input sections, 2-19, 2-22 command LDF scope, 3-19 DATA64 qualifier, 3-62 $COMMAND_LINE_LINK_AGAINST LDF data placement, 2-59 macro, 3-31 data sections, exception handling, 2-24 $COMMAND_LINE_OBJECTS LDF macro, .dat files, initialization data, A-3 3-14, 3-30 debugger, files, A-9 $COMMAND_LINE_OUTPUT_DIRECTO declaring, macros, 3-32 RY LDF macro, 3-31 default .ldf file, 3-3, 3-4 $COMMAND_LINE_OUTPUT_DIRECTO DEFAULT_OVERLAY () LDF command, RY macro, 2-61 3-71 $COMMAND_LINE_OUTPUT_FILE LDF DEFINED() LDF operator, 3-26 macro, 3-15, 3-31 directories, supported by linker, 2-48 $COMMAND_LINE_OUTPUT_FILE disassembly macro, 2-61 library member, B-3 commands, LDF, 5-29 using archiver, B-3 commands, See LDF commands using dumper, B-3 comma-separated option, 2-58 .dlb files comments in .ldf files, 3-18 defined, A-6 common memory, 3-39, 5-53 extension convention, 2-47 COMMON_MEMORY{} LDF command, symbol name encryption, 6-11 3-38, 5-39, 5-53 DMA accessing external memory, 3-46 compiler source files (.c .cc), 1-3 DMAONLY memory segment qualifier, 3-46 constdata input section, 2-22 DM qualifier, 3-62 constructors, 2-16, 2-20 -dnv archiver switch, 6-15 converting .doj files, 2-47, A-5 library members to source code, B-3 -Dprocessor (target architecture) linker switch, out-of-range short calls and jumps, 2-60 2-53 cplb_code input section, 2-22 dumper, in code disassembly, B-3 ___cplb_ctrl configuration variable, 3-9 DWARF reference information, A-10 cplb_data input section, 2-22 .dxe files Create LDF wizard, 4-4 data extraction, B-1, B-5 ctor0-4 input sections, 2-19 described, 1-7, 2-3, 2-6 ctor input section, 2-19, 2-22 extension conventions, 2-47 ctorl input section, 2-24 linker output files described, A-6 ___ctor_NULL_marker symbol, 2-18, 2-20, 3-42 E customer support, xxii .edt input section, 2-14, 2-19 -ek (no elimination) linker switch, 2-57 D -d archiver switch, 6-15

VisualDSP++ 5.0 Linker and Utilites Manual I-3

Index

elfar.exe encryption about, 6-1 constraints, 6-12 command-line reference, 6-14 symbol names in libraries, 6-11 constraints, 6-17 end address, memory segment, 3-47, 4-30 elfdump.exe utility, B-1 .end label ELF file, B-5 in assembly code, 1-5 contents, B-1, B-5 specifying function boundary, 3-39 ELF file dumper END() LDF identifier, 3-47 about, B-1 entry address command-line switches, B-1, B-5 ENTRY() command, 3-40 dumping contents of an output section, 2-13, global start symbol, 2-34 B-1, B-5 multiprocessor system, 2-35 extracting data, B-1 setting, 2-35 overlay library files, B-4 using the -entry switch, 2-58 reference information, A-10 -entry (entry address) linker switch, 2-58 used by Expert Linker, 4-34 ENTRY() LDF command, 3-40 elfloader.exe loader utility, 1-10 errata workaround, 2-63 ELF patch utility, B-5 errors, linker, 2-10 elfspl21k.exe splitter utility, 1-10 -es (eliminate listed sections) linker switch, 2-58 ELIMINATE() LDF command, 3-39 -ev (eliminate unused symbols, verbose) linker ELIMINATE_SECTIONS() LDF command, switch, 2-58 3-40 exclude_file archiver command-line parameter, elimination 6-15 enabling, 3-39, 3-42 executable files, 1-7, A-6 not applied to section, 2-57 post-processing, 2-61 restricting to named input sections, 2-58 EXPAND() LDF command, 3-58 specifying properties, 4-52 unused symbols, 2-58

I-4 VisualDSP++ 5.0 Linker and Utilites Manual

Index

Expert Linker overlay properties, 4-65 about, 4-1 overlays, 4-32, 5-2 adding input sections, 4-11 overview, 2-9, 4-2 adding LDF macros, 4-11 packing properties, 4-61 adding object files, 4-11 performance optimization, 5-2 adding output section to memory segment, processor properties, 4-49 4-20 profiling object sections, 4-37 adding overlay, 4-20 properties for overlays, 4-50 adding shared memory, 4-20 removing LDF macro, 4-11 adding shared memory segments, 4-42 resize cursor, 4-26 alignment properties, 4-63 shared memory properties, 4-70 allocating for heap, 4-67 sorting objects, 4-15 allocating for stack, 4-67 specifying new memory segments, 4-20 choosing initialization qualifier, 4-59 stack properties, 4-67 color selection, 4-13 symbols properties, 4-53 context menu, 4-19 viewing icons and colors, 4-13 deleting objects, 4-11 viewing section contents, 4-34 deleting selected object, 4-20 viewing symbols, 4-36 displaying global properties, 4-12 window panes, 4-9 displaying section contents, 4-12 expressions, in .ldf files, 3-20 elimination properties, 4-52 external execution packing, 3-53 expanding items, 4-20 external memory expanding LDF macros, 4-11 access, 3-46 global properties, 4-48 mapped explicitly into, 2-23 heap properties, 4-67 SHARC processors, 2-27, 2-29 icons, 4-13 TigerSHARC processors, 2-31 Input Sections pane, 4-10 extracting, data from ELF executable files, B-1, invalid memory segments, 4-17 B-5 launching, 4-3 Legend dialog box, 4-13 F mapping memory sections in, 4-12 memory management, 5-29 FALSE keyword, 3-23 memory management functions, 5-1 file extension conventions, 2-47 memory map graphical view, 4-22 @filename linker switch, 2-53 Memory Map pane, 4-16 memory segment properties, 4-57 multiprocessing tasks, 4-42 multiprocessor properties, 4-49 object properties, 4-47 output sections properties, 4-58

VisualDSP++ 5.0 Linker and Utilites Manual I-5

Index

file types .gdt, .gdtl input sections, 2-14, 2-19 .asm (assembly source files), A-3 global build, A-5 LDF file scope, 3-19 C/C++ source files, A-2 string, 2-44 .dat (assembly initialization data), A-3 symbol, 2-34 debugger, A-9 zero-initialized data, 2-15, 2-20, 2-23 default .ldf, 3-3, 3-4 Global Properties dialog box .dlb (library) General tab, 4-48 .doj (assembler object), A-5 PLIT tab, 4-51 .dxe, A-6 viewing, 4-21 executable, A-6 formats, A-1 H input format, A-2 .ldr (ASCII-format), A-9 hardware revision, building, 2-63 .ldr (hex format), A-6 he, 6-15 linker command-line (.txt), 2-47, A-5 header (.h) files, A-4 object, 2-49 heap output, 1-7 graphic representation, 4-67 .ovl, A-6 input section, 2-23 reference information, A-10 managing in memory, 4-67 .sm, A-6 hex-format files, .ldr format, A-6 .txt, A-5 -h (-help) switch, 2-59, 7-16 .xml, A-6 FILL() LDF command, 3-40, 3-69 I fill values, setting, 4-63 icons filter Expert Linker, 4-13 expression (optional), 3-65 unmapped icon, 4-12 operation, 2-43, 3-65 IDDE, See integrated development and FIRST_FIT LDF identifier, 3-71 debugging environment -flags-meminit linker switch, 2-58 IDDE_ARGS option, 2-44 -flags-pp linker switch, 2-59 -i filename archiver switch, 6-16 FORCE_CONTIGUITY LDF command, -IgnoreSection SectionName switch, 7-16 3-72 INCLUDE() LDF command, 3-40 fragmented memory, filling in, 2-59 individual placement, 2-59 .frt input section, 2-14, 2-19 initialization full trace, 2-64 flag, 7-8 initialization qualifier, choosing in Expert G Linker, 4-59 gap, inserting into memory segment, 4-31

I-6 VisualDSP++ 5.0 Linker and Utilites Manual

Index

initialization stream INPUT_SECTIONS() statement, 3-16 generated from memory initializer, 7-2 internal memory inserting executable file into, 7-16 Blackfin processors, 2-32 start address, 7-15 in Expert Linker, 4-31 structure, 7-5 SHARC processors, 2-28 -Init Initcode.dxe switch, 7-16 TigerSHARC processors, 2-30 ___inits symbol name, 7-15 inter-overlay calls, 5-27 InputFile.dxe switch, 7-17 inter-processor calls, 5-28 input files interrupt latch registers, 2-16 callback input executable file, 7-3 -ip (individual placement) linker switch, 2-59 primary input file, 7-3 input/output sections, contents of, 4-33 J INPUT_SECTION_ALIGN() LDF command, 3-40 -jcs2l (convert out-of-range short calls) linker input sections switch, 2-60 adding, 4-10 jumps, converting, 2-60 contiguity, 4-60 directives, 1-4 K names, 2-12 -keep (keep unused symbols) linker switch, 2-60 source code, 1-3 KEEP() LDF command, 3-42 viewing, 4-10 KEEP_SECTIONS() LDF command, 3-42 with corresponding output sections and keywords, 3-22, 3-23 memory segments, 2-13 INPUT_SECTIONS() LDF command, 3-64 Input Sections pane L displayed, 4-10 ___l1_code_cache guard symbol, 3-10 menu selections, 4-10 L1_DATA_A input section, 2-22 INPUT_SECTIONS_PIN_EXCLUSIVE LDF L1_DATA_B input section, 2-22 command, 3-67 ___l1_data_cache_a guard symbol, 3-10 INPUT_SECTIONS_PIN LDF command, ___l1_data_cache_b guard symbol, 3-10 3-67 LDF advanced commands, about, 5-29

VisualDSP++ 5.0 Linker and Utilites Manual I-7

Index

LDF commands .ldf files about, 2-4, 3-36 advanced commands in, 5-29 ALIGN(), 3-37 commands in, 2-4, 3-36 ARCHITECTURE(), 3-38 commenting, 3-18 COMMON_MEMORY{}, 3-38, 5-53 creating in Expert Linker, 4-4 ELIMINATE(), 3-39 default, 2-11 ELIMINATE_SECTIONS(), 3-40 defined, A-4 ENTRY(), 3-40 expressions in, 3-20 EXPAND(), 3-58 expression syntax, 3-20 FILL(), 3-69 extension conventions, 2-47 INCLUDE(), 3-40 generated by Blackfin processors, 3-3 INPUT_SECTION_ALIGN(), 3-40 keywords, 3-21, 3-22 INPUT_SECTIONS(), 3-64 main features, 3-13 KEEP(), 3-42 mapping output sections to memory segment, KEEP_SECTIONS(), 3-42 2-5 LINK_AGAINST(), 3-42 miscellaneous keywords used in, 3-23 MAP(), 3-43 operators, 3-23 MASTERS(), 5-53, 5-54 purpose, 2-6 MEMORY{}, 3-44, 5-46 scope, 3-19 MPMEMORY{}, 3-47, 5-45 specifying memory segment width, 2-5 OVERLAY_GROUP{}, 3-48, 5-29 structure of, 3-18 OVERLAY_INPUT{}, 3-70 used to map code/data to specific memory PACKING(), 3-48 segments, 3-3 PLIT{}, 3-69, 5-34 .ldf files, See LDF commands PROCESSOR{}, 3-54 LDF macros RESERVE(), 3-56 about, 3-29 RESOLVE(), 3-59 adding, 4-10 SEARCH_DIR(), 3-60 built-in, 3-30 SECTIONS{}, 3-61 command-line input, 3-32 SHARED_MEMORY{}, 3-72, 5-47 expanding, 4-11 __MEMINIT__, 3-36 predefined, 3-33 removing, 4-11 __SILICON_REVISION__, 3-36 user-declared, 3-32 using to partition code/data between processors, 5-40 __VERSION__, 3-35 __VERSIONNUM__, 3-35 __VISUALDSPVERSION__, 3-33

I-8 VisualDSP++ 5.0 Linker and Utilites Manual

Index

LDF operators library routines, accessing, 6-5 about, 3-23 LINK_AGAINST() LDF command, 3-42, ABSOLUTE(), 3-23 5-48, 5-50 ADDR(), 3-24 linker DEFINED(), 3-26 about, 2-2 location counter, 3-29 command-line files (.txt), A-5 MEMORY_END(), 3-26 command-line syntax, 2-45 MEMORY_SIZEOF(), 3-27 defined, 1-2 MEMORY_START(), 3-27 describing the target, 2-11 SIZEOF(), 3-28 error messages, 2-10 ld linker (UNIX), B-7 executable files, A-6 .ldr files file duplications by, 5-59 ASCII-format, A-9 file name conventions, 2-48 hex-format, A-6 generating PLIT constants, 5-36 splitter output, A-9 linking object files, 2-49 leaf functions, 5-59 mapping by attributes, 5-43 legends, Expert Linker, 4-11 mapping using object archive, 5-44 LENGTH() LDF identifier, 3-47 memory map files (.xml), A-6 length_symbol symbol, 3-56 options, 2-4 ___lib_end_of_heap_descriptions symbol, 3-42 output files, 1-7, A-6 lib_file archiver command-line parameter, 6-15 outputs, 1-7, 2-3, 2-6 librarian, VisualDSP++, 6-1 overlay constants generated by, 5-9 $LIBRARIES library and object file list, 3-13 running from command line, 2-45 library, symbol name encryption, 6-11 running from VisualDSP++, 2-7 library files (.dlb) switches, 2-49 about warning messages, 2-10 adding, 4-10 Linker Description Files defined, A-6 overview, 3-1 searchable, 6-2 linker.exe, 1-2 library members linker-generated constants, 5-7, 5-15, 5-16 converting to source code, B-3 linker-generated overlay constants, 5-9 linking into executable program, 6-1 linker macros, 3-29, 3-30

VisualDSP++ 5.0 Linker and Utilites Manual I-9

Index

linker switches -Wnumber (warning suppression), 2-57 -Darchitecture, 2-53 -Wwarn num (override error message), 2-57 -Dprocessor, 2-53 -xref filename, 2-65 -e (eliminate), 2-57 linking -ek secName, 2-57 about, 2-3 -entry, 2-58 controlling, 2-4 -es secName, 2-58 environment, 2-7 -ev, 2-58 file with large uninitialized variables, C-4, @filename, 2-53 D-4, E-4 -flags-meminit, 2-58 file with large zero-initialized variables, C-4, -flags-pp, 2-58 D-4, E-4 -h (help), 2-59 multiprocessor SHARC systems, D-6 -i (include search directory), 2-59 multiprocessor systems, 5-39 -ip (individual placement), 2-59 multiprocessor TigerSHARC systems, E-6 -jcs2l, 2-60 process rules, 2-5 -keep symbolName, 2-61 single-processor Blackfin system, C-2 -M, 2-54 single-processor SHARC system, D-2 -Map filename, 2-55 single-processor TigerSHARC system, E-2 -MDmacro, 2-55 with attributes, 2-42 -meminit, 2-61 Link page, setting linker options, 2-7 -MM, 2-54 link target, 2-11 -MUDmacro, 2-56 loader -nonmemcheck, 2-61 creating bootloadable image, 1-10, 1-11 -od directory, 2-61 hex-format files, A-6 -o filename (output file), 2-61 output files (.ldr), A-6 -pp, 2-62 location counter, definition of, 3-29 -proc processor, 2-62 -S, 2-56 M -save-temps, 2-63 -si-revision version (silicon revision), 2-63 macros -sp (skip preprocessing), 2-64 LDF, 3-29 -s (strips all symbols), 2-62 preprocessor, 3-30 -T filename, 2-56 undefining, 2-56 -t (trace), 2-64 user-declared, 3-32 -tx (full trace), 2-64 main function, 2-44 -version (display version), 2-64 -Map (filename) linker switch, 2-55 -v (verbose), 2-64 map file (.xml), 2-49, 2-55, 3-43 -warnonce, 2-64 MAP() LDF command, 3-43 -Werror num (override warning message), 2-57

I-10 VisualDSP++ 5.0 Linker and Utilites Manual

Index

mapping MEMORY_END() LDF operator, 3-26 archives, 5-53, 5-54 memory initializer by attributes, 5-43 about, 7-2 by section name, 5-42 basic operations, 7-3 input sections to output sections, 4-12 command line switches, 7-14 input section to several output sections, 3-67 extracting data from section, 7-18 into memory sections in common memory, function of, 7-1 5-55, 5-58 ___inits default symbol name, 7-15 using archive or library, 5-44 invoking, 2-61, 7-10, 7-12 -M archiver switch, 6-16 .ldf file preparation, 7-7 master processor, 3-38, 5-53, 5-54, 5-56 NO-BOOT mode, 7-2 MASTERS() LDF command, 5-53, 5-54 output file, 7-18 -M (dependency check and output) linker passing comma-separated option to, 2-58 switch, 2-54 primary input file, 7-17 -MDmacro (macro value) linker switch, 2-56 section initialization flag, 7-11 mem_argv input section, 2-19 switches, See memory initializer switches MEM_ARGV memory section, 2-44 when to use, 7-7 __MEMINIT__ LDF macro, 3-36 memory initializer switches -meminit linker switch, 2-61 -BeginInit InitSymbol, 7-15 .meminit memory section, serving memory -h (help), 7-16 initializer, 7-8 -IgnoreSection SectionName, 7-16 memory -Init Initcode.dxe, 7-16 allocation, 2-12 InputFile.dxe, 7-17 architecture representation, 2-11 -NoAuto, 7-17 Blackfin processor, 2-32 -NoErase, 7-17 common, 3-39, 5-53 -o OutputFile.dxe, 7-18 external multiprocessor space -Section SectionName, 7-18 (TigerSHARC), 2-31 -v (-verbose), 7-18 initialization, 3-62, 7-6 memory interface, width (bits), 3-47 internal, 2-30 MEMORY{} LDF command managing heap/stack, 4-67 described, 5-46 map files, A-6 .ldf file component, 3-15 mapping, 4-12, 5-42 segment_declaration, 3-45 multiprocessor (TigerSHARC), 2-31 syntax diagram, 3-44 overlays, 5-4, 5-5 using in an .ldf file, 2-32 partitions, 4-16 writing, 2-12 segments, 2-12, 3-47, 4-16 SHARC processor, 2-27 TigerSHARC processor, 2-30 types, 2-12, 3-46

VisualDSP++ 5.0 Linker and Utilites Manual I-11

Index

memory map MPMEMORY{} LDF command, 3-47, 5-45 adding memory segments to, 4-29 -MUDmacro (undefine macro) linker switch, generating, 2-55 2-56 graphical view, 4-23 multicore highlighted objects in, 4-26 applications, 2-35, 3-56 post-link view, 4-26 systems with shared SDRAM, 2-24 pre-link view, 4-26 multiprocessor specifying, 2-12 applications, 2-35, 3-56 tree view, 4-21 linking commands, 5-39 viewing, 4-17 memory, 2-31 Memory Map pane multiprocessor systems context menu, 4-19 bus lock in, D-12 described, 4-17 code/data placement in, 5-40 overlays, 4-32 heterogeneous, 5-39 right-click menu, 4-19 homogeneous, 5-39 zooming in/out, 4-28 linking, 5-39 memory sections managing processor order, 4-50 Blackfin processors, 2-21 processor physical memory offset, 5-45 SHARC processors, 2-14 semaphores, D-12 TigerSHARC processors, 2-19 shared memory, 5-47 memory segments about, 1-3 N adding, 4-29 adding to memory map, 4-29 -NoAuto switch, 7-17 changing size of, 4-25 -NoErase switch, 7-17 inserting a gap, 4-31 NOFORCE_CONTIGUITY LDF command, invalid, 4-17 3-72 MEMORY{} command, 4-16 NO_INIT qualifier, 3-63, C-4, D-4, E-4 new, 4-20, 4-29 -nomemcheck linker switch, 2-61 rules, 2-5 noncache_code input section, 2-23 size, 4-22 specifying properties, 4-57 O start address, 4-22 object files MEMORY_SIZEOF() LDF operator, 3-27 adding, 4-10 memory space, 4-30 explained, 1-3 MEMORY_START() LDF operator, 3-27 linking into executable, 2-3 -MM archiver switch, 6-16 object properties, managing with Expert Linker, -MM (dependency check, output and build) 4-47 linker switch, 2-54 modify register, 5-17

I-12 VisualDSP++ 5.0 Linker and Utilites Manual

Index

objects overlay library files, B-4 deleting, 4-11 overlay manager sorting, 4-11, 4-15 about, 5-4, 5-6, 5-7 $OBJECTS LDF macro, 3-13 assembly code, 5-36 obj_file archiver command-line parameter, 6-15 constants, 5-15 -od (output directory) linker switch, 2-61 major functions, 5-7 offset, processor physical memory, 5-45 performance summary, 5-20 -o filename linker switch, 2-61 placing constants, 5-17 -o OutputFile.dxe switch, 7-18 PLIT table, 5-12 ## operator, 1-8 routines, 5-5 operators, .ldf file, 3-23 storing overlay ID, 5-20 output directory, specifying, 2-61 OVERLAY_OUTPUT() LDF command, 3-71 OUTPUT() LDF command, 3-15, 4-16 overlays output section qualifiers, 3-63 address, 5-9, 5-16 output sections constants, 5-9, 5-15 about, 2-4 debugging, 5-7 dumping, 2-13 dumping library files, B-4 mapped contiguously, 4-60 grouped, 5-30, 5-33 new, 4-20 live address, 4-66 overflow, 4-59 live space, 4-32 rules, 2-5 loading and executing, 5-21 specifying properties, 4-58 loading instructions with PLIT, 5-38 _ov_endaddress_# overlay constant, 5-9, 5-26 managing properties, 4-65 _ov_end breakpoint, 5-7, 5-8 memory, 5-4, 5-5 overlay Memory Map pane, 4-32 ALL_FIT algorithm, 4-65 multiple, 4-32 file, producing, 3-71 numbering, 5-27 live space, 4-32 reducing overhead, 5-21 new, 4-20 run address, 4-66 run space, 4-32 run space, 4-32 OVERLAY_GROUP{} LDF command, 3-48, size, 4-66 5-29 special symbols, 5-26 overlay ID, storing, 5-20 ungrouped, 5-30, 5-31 OVERLAY_ID LDF identifier, 3-71 word size, 5-9, 5-16 OVERLAY_INPUT{} LDF command ov_id_loaded buffer, 5-20 DEFAULT_OVERLAY() portion, 3-71 described, 3-70

VisualDSP++ 5.0 Linker and Utilites Manual I-13

Index

.ovl files PLIT described, A-6 about, 5-10 diagram, 1-7, 2-3, 2-6 allocating space for, 5-37 dumping, B-4 executing user-defined code, 5-10 extracting content from, B-4 overlay constants, 5-36 file conventions, 2-47 overlay management, 5-7 linker output, A-6 specifying overlay properties, 4-50 OVERLAY_INPUT{} command used in, summary, 5-38 3-71 syntax, 5-35 viewing content, B-4 PLIT{} LDF command _ov_runtimestartaddress_# overlay constant, about, 5-34 5-9, 5-27 in SECTIONS{}, 3-69 _ov_size_# overlay constant, 5-9, 5-26 instruction qualifier, 5-35 _ov_startaddress_# overlay constant, 5-9, 5-26 LDF input section, 3-69 _ov_start breakpoint, 5-8 overview, 3-53 _ov_word_size_live_# overlay constant, 5-9, PLIT_SYMBOL_ADDRESS, 5-36 5-27 PLIT_SYMBOL_OVERLAYID, 5-36 _ov_word_size_run_# overlay constant, 5-9, syntax described, 5-34 5-26 .plit output section, 5-36 PLIT_SYMBOL_ADDRESS, 5-36 P PLIT_SYMBOL constants, 5-38 PLIT_SYMBOL_OVERLAYID, 5-36 packing PM qualifier, 3-62 data, 3-48 -pp (end after preprocessing) linker switch, 2-62 DMA, 3-50 pp.exe preprocessor, 1-8 external execution, 3-52 #pragma section, 5-43 header files, 3-50 preprocessor in SHARC processors, 3-50 compiler, 1-9 overlay format, 3-51 linker and assembler commands, 1-8 properties, specifying, 4-61 macros, 3-30 with PACKING() LDF command, 3-48 running from linker, 2-62 PACKING() LDF command, 3-48 procedure linkage table (PLIT) -p archiver switch, 6-16 see also PLIT partial linker, B-7 about PLIT{} command, 3-53, 5-34 partially-linked relocatable object file, B-7 summary, 5-10 PGO, See profile-guided optimization using, 5-25 pinning objects to output section, 4-20, 4-58 plinker command-line tool, B-6

I-14 VisualDSP++ 5.0 Linker and Utilites Manual

Index

PROCESSOR{} LDF command RTL routine, performing memory initialization, declaring a processor and its related link 7-2, 7-6 information, 3-54 .rtti input section, 2-14, 2-19 .ldf file component, 3-15 run-time initialization linking projects on multiprocessor/multicore data (in bsz_init section), 2-20, 2-23 Blackfin architectures, 3-56 qualifiers, 3-63 syntax, 3-54 type_qualifier, 3-62 Processor Properties dialog box, 4-49 RUNTIME_INIT qualifier, 7-7 processors defined, 3-63 common memory, 5-53 example, 7-9 selection of, 2-53, 2-62 -IgnoreSection switch, 7-16 sharing memory, 5-54 --NoAuto switch, 7-17 silicon revision of, 2-63 -Section switch, 7-18 specifying properties of, 4-49 run-time libraries, built using attributes, 5-44 -proc (target processor) linker switch, 2-62 profile-guided optimization (PGO) S IDDE_ARGS option, 2-43 linker support for, 2-43 -s archiver switch, 6-16 PROGBITS qualifier, 7-4 -save-temps linker switch, 2-63 program input section, 2-19, 2-21 SDRAM, enabled, 2-24 project builds, linker, 2-7 sdram0_bank input section, 2-23 Project Options dialog box, 2-7 sdram0 input section, 2-23 PROM, specified by TYPE() command, 3-46 sdram_bcz input section, 2-24 Properties dialog box, viewing, 4-21 sdram_shared input section, 2-24 -pva archiver switch, 6-9, 6-16 SEARCH_DIR() directory paths, 3-13 -pv archiver switch, 6-9, 6-16 SEARCH_DIR() LDF command, 3-60 sec_rth input section, 2-14 section R contents, 4-12 RAM, specified by TYPE() command, 3-46 formats, selecting, 4-35 RAM selection, 4-30 input, 2-13 -r archiver switch, 6-16 .SECTION assembly directive, 1-4 reflective semaphores, D-12 section contents, viewing, 4-33 removing, LDF macro, 4-11 section mapping RESERVE() LDF command, 3-56 Blackfin processors, 2-21 -reserve-null linker switch, 2-62 SHARC processors, 2-14 resize cursor, 4-26 TigerSHARC processors, 2-19 RESOLVE() LDF command, 3-43, 3-59 section_name qualifier, 3-62 ROM, specified by TYPE() command, 3-46 section pragma, 5-43 ROM selection, 4-30 -Section SectionName switch, 7-18

VisualDSP++ 5.0 Linker and Utilites Manual I-15

Index

SECTIONS{} LDF command, 2-36, 3-16, 3-61 SHARC memory sections .ldf file component, 3-16 section mapping, 2-14 specifying placement of code/data in physical .bss input section, 2-15 memory, 3-16 exception handling data sections, 2-18 Linker Description Files .gdt .gdtl .frt .frtl .cht .chtl .edt .edtl, 2-18 see also .ldf files .rtti input section, 2-15 seg_argv input section, 2-14 seg_argv input section, 2-16 seg_ctdm input section, 2-14 seg_ctdm input section, 2-16 seg_ctdml input section, 2-14 seg_ctdml memory section, 2-18 seg_dmda input section, 2-14 seg_dmda memory section, 2-16 seg_ext_code input section, 2-14, 2-15 seg_ext_dmda input section, 2-18 seg_ext_data input section, 2-15 seg_ext_nwco input section, 2-18 seg_flash input section, 2-15 seg_ext_pmda input section, 2-18 seg_heap input section, 2-15 seg_ext_swco input section, 2-18 seg_init_code input section, 2-14 seg_flash input section, 2-17 seg_init input section, 2-14 seg_heap input section, 2-17 segment declaration, 3-45 seg_init input section, 2-15 segment end address, 3-57 seg_int_code input section, 2-16 seg_pmco input section, 2-14 seg_pmco input section, 2-16 seg_pmda input section, 2-14 seg_pmda input section, 2-16 seg_sdram input section, 2-15 seg_rth input section, 2-15 seg_sram input section, 2-14 seg_sram input section, 2-17 seg_stak input section, 2-14 seg_stak input section, 2-17 seg_swco input section, 2-15 seg_vtbl input section, 2-17 seg_vtbl input section, 2-14 SHARC processors semaphores, reflective, D-12 basic .ldf file example, 3-11 .SEPARATE_MEM_SEGMENTS assembler broadcast space, D-12 directive, 2-41 external memory, 2-29 implementing reflective semaphores, D-12 internal memory, 2-28 LDF programming examples, D-1 memory architecture, 2-27 memory packing, 3-52 multiprocessor (MP) systems, D-12 overlay packing format, 3-52 packing in, 3-50

I-16 VisualDSP++ 5.0 Linker and Utilites Manual

Index

shared memory special section name (.PLIT), 3-62 managing properties of, 4-70 splitter mapping objects into, 3-39, 5-53 ASCII-format files (.ldr), A-9 new, 4-20 generating non-bootable PROM image files, SHARC system, D-6 1-11 TigerSHARC system, E-6 output file (.ldr), A-8 used with multiprocessor systems, 5-47 SPORT data files, A-9 SHARED_MEMORY{} LDF command, 3-72, -sp (skip preprocessing) linker switch, 2-64 5-39, 5-47 -s (strip all symbols) linker switch, 2-62 short calls, converting, 2-60 -S (strip debug symbols) linker switch, 2-56 SHT_NOBITS stack keyword, 3-63, C-4, D-4, E-4 graphic representation, 4-67 section qualifier, C-4, D-4, E-4 input section, 2-23 silicon revision, selecting, 2-63 managing in memory, 4-67 __SILICON_REVISION__ LDF macro, 3-36 START() command, 3-45 -si-revision (silicon revision) linker switch, 2-63 start_symbol symbol, 3-56 size (hexadecimal), 4-30 SW qualifier, 3-62 SIZE() LDF command, 3-72 symbols SIZEOF() LDF operator, 3-28 adding, 4-53, 4-56 .sm files declaration, 3-13 described, A-6 deleting from resolve list, 4-56 diagram, 1-7, 2-3, 2-6 encryption of names, 6-11 file extension conventions, 2-47 manager, 5-7 linker output, A-6 managing properties of, 4-53 sort, objects, 4-11, 4-15 removing, 2-62, 4-53 source code, in input sections, 1-3 resolution, 4-54, 4-56 source files viewing, 4-20, 4-37, 4-53 assembly initialization data (.dat) files, A-3 C/C++, A-2 T command-line file (.txt), A-5 compiling into object files, 1-3 -t archiver switch, 6-7, 6-16 fixed-point data, A-3 target architecture (processor), 2-54, 3-13 header files, A-4 -T file (executable program placement) linker .ldf, A-4 switch, 2-56 preparing, 7-9

VisualDSP++ 5.0 Linker and Utilites Manual I-17

Index

TigerSHARC memory sections USE_CACHE configuration, 3-9 section mapping, 2-19 user-declared macros, 3-32 bsz_init input section, 2-20 utilities bsz input section, 2-20 archiver (elfar.exe), 6-1 ctor1 memory section, 2-21 producing relocatable object files, B-7 ctor input section, 2-20 data1 input section, 2-20 V data2 input section, 2-20 exception handling data sections, 2-21 -version (display version) .gdt .gdtl .frt .cht .edt .rtti data sections, 2-21 archiver switch, 6-16 mem_argv input section, 2-20 linker switch, 2-64 program input section, 2-20 version information vtbl input section, 2-21 built in with archiver, 6-7 TigerSHARC processors user-defined, 6-8 allocating buffers to different memory __VERSION__ LDF macro, 3-35 segments, 2-41 __VERSIONNUM__ LDF macro, 3-35 basic .ldf file example, 3-10 VisualDSP++ external memory, 2-31 archiver, 6-1 internal memory, 2-30 Assemble page, 2-8 LDF programming examples, E-1 Expert Linker, 4-2 memory architecture, 2-30 integrated development and debugging programming enhancement, 2-41 environment (IDDE), 2-2 -tnv archiver switch, 6-16 librarian, 6-1 tree view, memory map, 4-21 project builds, 2-7 TRUE keyword, 3-23 Project Options dialog box, 2-7, 2-8 .tst files, linker, A-5 running linker from, 2-7 -t (trace) linker switch, 2-64 setting assembler options, 2-8 -twc ver archiver switch, 6-16 setting options, 2-8 -tx filename archiver switch, 6-16 __VISUALDSPVERSION__ LDF macro, -tx (full trace) linker switch, 2-64 3-33 .txt files, linker command-line files, A-5 voldata input section, 2-22 TYPE() command, 3-46 vtbl input section, 2-19, 2-24 type-letter archiver command-line parameter, -v (verbose) 6-15 archiver switch, 6-16 linker switch, 2-64 MemInit switch, 7-18 U uninitialized variables, C-4, D-4, E-4 W UNIX linker (ld), B-7 unmapped object icon, 4-12 warnings, linker, 2-10 unpacking, data, 3-49

I-18 VisualDSP++ 5.0 Linker and Utilites Manual

Index

-warnonce (single symbol warning) linker X switch, 2-64 xmlmap2html.exe command-line utility, 2-55 -Werror num (override warning message) linker .xml map file switch, 2-57 description, A-6 width, memory segment, 4-30 generating, 2-55 WIDTH() command, 3-47 MAP filename command, 3-43 wildcard characters opening in Web browser, 2-55 in section names, 2-35, 3-36 XREF keyword, 3-23 specifying archive files, 6-16 XSLT, language for transforming XML using in archiver, 6-6 documents, 2-55 wizards, Create LDF, 4-4 -Wnnnn archiver switch, 6-16 -Wnumber (warning suppression) linker switch, Z 2-57 ZERO_INIT qualifier, 7-7 word width (number of bits), 3-47 defined, 3-63 -w (remove warning) archiver switch, 6-16 example, 7-9 -Wwarn num (override error message) linker -IgnoreSection switch, 7-16 switch, 2-57 --NoAuto switch, 7-17 -Section switch, 7-18

VisualDSP++ 5.0 Linker and Utilites Manual I-19